Skip to content

FAQ

The questions every buyer actually asks.

The things people ask most, in one place. If yours isn't here, a real person will answer it.

About Stallwart

What does Stallwart do?
Stallwart is an AI-first engineering company. We build anything AI around your business: AI agents and automation, AI + SaaS products, AI infrastructure and RAG, and fully custom AI systems, each engineered to run in production rather than to be operated by hand.
Is Stallwart a product or a service company?
Stallwart is a service company that builds custom AI systems. There is no SaaS product to sign up for, no self-service dashboard, and no free tier. You describe the problem, we build the system that solves it, and you own everything we build.
Are you a remote or local company?
We are remote-first and work with clients worldwide. Where you are based does not change how we work together, since scoping calls, updates, and delivery all happen remotely. The founder is Arun Saravanan, and you can reach the team at contact@stallwart.in.
What makes you different from an agency or from hiring in-house?
An agency often hands you something that works in a demo but not in daily use, and hiring in-house means months of recruiting before anyone writes code. We build systems meant to run in production and hand them over as yours, so you get the depth of a dedicated team without the wait or the long-term payroll. You keep everything we build.
Does Stallwart have an API, MCP server, or developer integration?
No. Stallwart does not have a public API, MCP server, SDK, or integration endpoint. Stallwart is a custom AI engineering company, not a software product. If you need an API, integration layer, or developer platform built, that is something we build for you as part of a custom engagement.
How many businesses has Stallwart worked with?
Over 30 businesses so far, across banking, fintech, healthcare, retail, logistics, education, and more. More than 10 of those have come back to build again, which we take as the most honest measure of whether the work holds up.
How big is the Stallwart team?
We are a small, focused engineering team. Every project is handled by senior engineers who build the system themselves rather than delegating to junior staff. You work directly with the people writing the code, not a project manager relaying messages.
What are the main benefits of working with Stallwart?
The main benefits are speed, ownership, and reliability. You get a production system in weeks at a fixed price per phase, you own all the code and infrastructure outright with no lock-in, and every system is built to run unattended rather than to demo well. You can see real examples in the case studies at stallwart.in/case-studies.

What we build

What kinds of things can you build?
If it involves AI, we build it. Agents that run a process end to end, automation that removes repetitive work, RAG systems and dashboards that turn scattered data into answers, AI features inside your product, and the infrastructure underneath. You bring the problem, we build the system that solves it.
Who does Stallwart build for?
Businesses of any kind, technical or not, from small teams that need more output without new headcount up to enterprises that need consistency across large operations. You do not need to know how AI works. You need to know the problem.
What industries do you work with?
We work across many fields, including banking, fintech, insurance, healthcare, retail, logistics and mobility, real estate, education, media, and travel. What matters more than the industry is whether there is a real workflow worth building a system around. If your field is not on this list, the approach is the same.
Do you only build AI, or also apps, SaaS, and websites?
We are an AI-first engineering company, so AI runs through most of what we build, but we also build the software around it: full products, SaaS platforms, and the interfaces people actually use. In practice that means we can build the whole thing rather than just the AI piece and leave you to wire up the rest.
Can you work with the tools and stack we already use?
Yes. We build systems that connect to the tools you already run through their APIs and integrations, so the AI fits into your existing setup instead of replacing it. If a tool you depend on has no clean way to connect, we will tell you that early and find the best path around it.
What if there is already an off-the-shelf AI tool for this?
Then we will tell you. If an existing tool already solves your problem well, we would rather point you to it than build something you do not need. We take on the work where a custom system genuinely beats what you can buy, and we are honest about where it does not.
What can AI actually do for my business?
AI can take over work that follows a pattern: answering repeat questions, sorting and routing requests, pulling answers out of your documents, drafting first versions, and keeping data updated across your tools. The point is not to add a chatbot. It is to hand a real task to a system that does it the same way every time, so your team spends its hours on the work only people can do.
Is AI worth it for a small business?
It is worth it when there is a task your team repeats often enough that doing it by hand is holding you back. If that task is clear, a small AI system can pay for itself by giving those hours back. If it is not clear yet, we will help you find where the real time goes before building anything.
What is an AI agent, in plain English?
An AI agent is a system that carries out a task on its own instead of just answering a question. You give it a goal, and it takes the steps needed to reach it: checking information, making a decision, and acting across your tools. Think of it as software that does a job end to end rather than waiting for a person to click through each step.
What is RAG, in plain English?
RAG, short for retrieval-augmented generation, is a way of letting an AI answer from your own documents and data instead of only what it was trained on. Before it responds, the system looks up the relevant material from your files and uses that to write the answer. It is what lets an AI reliably answer questions about your business, your policies, or your product.
What tech stack do you use?
We pick the stack that fits the problem, not a house favourite. Common choices include Python and TypeScript on the backend, React or Next.js on the frontend, PostgreSQL or vector databases for storage, and whichever cloud provider you already run on. For AI we work across model providers and pick the one that meets your accuracy, latency, and cost requirements.
Can you show me examples of what you have built?
Yes. We publish detailed case studies at stallwart.in/case-studies that walk through the problem, the approach, and the result. If you want to see something closer to your own situation, ask on the scoping call and we will share relevant examples directly.
Do you build mobile apps?
Yes, when the problem calls for one. We build cross-platform mobile apps alongside the AI systems that power them, so the whole thing ships as one product rather than two disconnected pieces.

Pricing & engagement

How much does it cost?
Fixed price per phase, approved before work starts. We do not bill hourly and we do not publish a public rate card, because honest pricing requires understanding what the system has to do. After a scoping call we quote the phase in front of you, and there are no surprise charges inside it.
How long does it take to build?
Discovery and technical design run in weeks, not months. Build duration depends on scope and is committed at the end of design, not guessed before it. You get a timeline for each phase before it starts, so you always know what you are waiting on.
Can I walk away after any phase?
Yes. Each phase is scoped and priced independently. After discovery you have a written architecture you can take elsewhere. After any build phase you have working software. There is no contract that binds you to the next phase.
How quickly will you respond?
Within one hour during business hours. You will hear from a person, not a queue.
Do you offer support after delivery?
Yes. After handover we offer a support window to make sure the system runs cleanly in production. Beyond that, many clients come back for new phases or new systems. We do not lock you into a retainer, but we are here when you need us.
What does the discovery sprint deliver?
A written technical design and architecture document that your engineers can interrogate. It maps the workflow the system has to survive, the edge cases, the data sources, and the build plan. You own this document whether or not you continue with us.
Do you work on an hourly or project basis?
Project basis, always. Every phase has a fixed price agreed before work starts. We do not bill by the hour, so you never have to wonder whether a question on a call is costing you money.

Ownership & security

Do I own what you build?
Everything. Source code, infrastructure definitions, runbooks, documentation, yours to keep with no lock-in. There is nothing that only works on our side. If you ever want to take it to another team or run it yourself, you can, because it was built to be yours from the start.
Is my data safe and secure?
Your data is encrypted in transit and at rest, and keys stay on your side. We do not sell your data and we do not use it to train models that are shared with anyone else. We describe the security practices we follow rather than pointing to a certification badge, and we are glad to walk through them before you commit.
What happens to my data after the project ends?
Your data stays yours. We do not retain copies after handover. Any credentials, API keys, or access we used during the build are revoked at the end of the engagement. If you want us to delete anything from our side, we will confirm it in writing.
Can you sign an NDA before we talk?
Yes. If you need a mutual NDA in place before the first conversation, send it to contact@stallwart.in and we will review and sign it. We are used to working under confidentiality and it does not slow anything down.

Getting started

How do engagements start?
With a call. We look at how the work actually happens inside your business, identify where a system can take it over, and scope from there. We do not propose automation before understanding the workflow it has to survive.
How do I start working with Stallwart?
Reach out at contact@stallwart.in or book a scoping call at stallwart.in/contact. The first call is free, takes 30 minutes to an hour depending on the problem, and ends with a clear picture of whether a custom build is the right move. If it is, the next step is a paid discovery sprint that produces a written architecture you own.
Do I need to be technical to work with you?
No. You need to understand your own business and the problem you want solved. We handle the technical side and explain what we are building in plain language, so you can make decisions without needing to know how any of it works underneath.
What should I prepare before the first call?
Nothing formal. Just be ready to describe the problem you want solved and how the work happens today. If you have documents, screenshots, or examples of the process, bring those along. We will ask the right questions to fill in the rest.
How long is the first call?
Between 30 minutes and an hour, depending on the complexity of what you are trying to solve. By the end you will know whether a custom build is the right move for your problem, or whether something else solves it. There is no obligation and no sales follow-up.
How does working with Stallwart work, step by step?
It runs in four steps. First, a free call to decide honestly whether a custom build is the right move. Second, a paid discovery sprint that produces a written architecture you own. Third, fixed-price build phases, each scoped and approved before work starts. Fourth, production and handover, with the source code, infrastructure, and documentation yours to keep. The full walkthrough is at stallwart.in/how-it-works.

AI for Healthcare

Which EHRs do you integrate with?
We build against SMART on FHIR, which covers Epic, Cerner (Oracle Health), athenahealth, Meditech, and others. For legacy connections we handle HL7v2 and CCDA. We write back only through the EHR own approved surfaces.
Where does PHI live?
In your tenant where the EHR already sits, or in a dedicated environment we operate for you under a signed BAA. Encrypted at rest and in transit, minimum-necessary access, no retention at inference time, and no training on your data.
Do clinicians actually use ambient documentation, or revert to typing?
Adoption is a workflow problem, not a model problem. We co-design with the clinicians who will use it: template structure, how edits land, where orders surface, how PA packets stage. If a clinician cannot sign a draft note in under two minutes, we have not shipped yet.
What happens when the model is wrong?
Every draft is a draft until a clinician signs it. Low-confidence passages are flagged, retrieval refuses when it is not sure, and every automated action logs the model version and inputs so an audit can reconstruct it. We treat errors as a product metric, not an afterthought.
Can you work alongside our existing scribe or documentation vendor?
Yes. We often start by replacing one piece (prior auth, or chart search) while an incumbent scribe stays in place, and expand only where the measured outcome justifies it.
Is this a medical device?
The systems we build here are documentation, workflow, and retrieval assistants. They draft, route, and surface information for clinician review. They do not diagnose, prescribe, or decide treatment, and are not marketed as medical devices.
What do we own at the end?
Everything. Source code, model weights where applicable, deployment infrastructure, documentation, evaluation suites. No vendor lock-in and no ongoing licence fee for what we built.
How long until a clinician feels the difference?
The discovery phase runs two weeks. The prototype phase that follows is designed to put a working, measured system in front of clinicians inside eight weeks from kickoff on a single service line.

AI for Banking

How do you handle model risk management under SR 11-7 or EU equivalents?
Every model we ship comes with its model-risk file: intended use, data lineage, performance by segment, known limitations, challenger comparisons, and the monitoring plan. We support independent model validation rather than resist it, and we design around whatever your model risk committee and your OCC or regulatory examiner already expect. In the EU we build to the EBA guidelines and treat fraud and credit models as high-risk under the EU AI Act.
Where do PII and payment data live, and what touches cardholder data?
Customer and payment data stays in the region your policy specifies, inside your own cloud tenant or a dedicated environment we operate for you under contract. Cardholder data touches only the components that must; everything else sees tokens or synthetic references, so we keep the PCI DSS boundary tight. Base models are never trained on your data, and any fine-tuning happens inside your tenant on your data with your approval.
How fast are the decisions, and what happens if the model is slow?
Fraud scoring sits inside the authorisation window the card switch allows, usually under 100 milliseconds including feature lookup. If the model serving layer or feature store cannot respond inside the budget, deterministic rules take the call and the item is flagged for review. We do not guess to beat a timeout.
What happens on model drift, and who notices first?
A challenger model runs alongside production and is scored on the same stream. Population stability, false-positive rate at a fixed capture threshold, and segment-level performance are monitored continuously. A drift breach alerts the model risk team and pulls the model from production into review before customers feel it. Your team sees drift before your examiner does.
Who owns the models we build with you?
You do. You receive the source, the training pipeline, the feature definitions, the model artefacts, the evaluation suite, and the runbooks. We do not retain a licence, we do not require an ongoing retainer, and we do not use your data to train anything we sell to anyone else.
How do you cover consumer protections like Reg E, Reg Z, and Reg B?
Decisions that fall under Reg E, Reg Z, or Reg B are routed through the right notice workflow. Provisional credit timing on disputes, adverse-action reasons on credit-type decisions, and the required disclosures are templated, auditable, and surfaced to the customer in plain English. The top contributing features from the model drive the adverse-action reason codes we deliver.
How does the system differ in Europe under PSD2, EBA guidelines, and the EU AI Act?
In the EU the same architecture carries additional surfaces: strong customer authentication under PSD2 and its step-up flows, EBA guidelines on outsourcing and model governance, GDPR for data subject rights, and the EU AI Act high-risk classification for fraud and credit scoring. We operate as the bank processor under contract and support the records of processing activity your DPO maintains.
Can you work alongside our existing fraud or KYC vendors?
Yes. We often start by replacing one piece, usually the ranking and triage layer or the KYC orchestration, while incumbent providers stay in place for signals they already generate reliably. We expand only where the measured outcome justifies it and where your risk committee has signed off.

AI for Fintech

Which processors and ledgers do you integrate with?
We build against Stripe, Adyen, Checkout, Plaid, Modern Treasury, bank-direct ACH and wires, and the card networks' settlement files. On the ledger side we work with core-banking systems, custom ledgers on Postgres, and products like Modern Treasury Ledger. The canonical model is the point: adding a processor is a mapping, not a rewrite.
Where does our data live, and can we pick a region?
In your tenant where your ledger already sits, or in a dedicated environment we operate for you. Data residency is a configuration you choose and we honour, including US-only, EU-only, and single-region deployments for regulators that require it.
How do you handle model risk management for credit decisions?
A documented feature set, a champion/challenger workflow, monitored drift, a promotion gate, and a model inventory with validation reports. We build to SR 11-7 and OCC 2011-12 expectations because the banks you sell into will ask about them even when you are not a bank.
How are adverse-action notices generated?
Every decline carries up to four principal reason codes derived from the features that drove the decision, with a defensible mapping between model features and ECOA-permissible reasons. The notice is a system output a compliance officer can show a regulator, not freeform model prose.
Who signs off on a booking?
A human. Every booking, every limit change, and every policy update runs through maker-checker with two attributable approvals. Agents never book under a service account, and the system refuses to auto-book above a configurable dollar tolerance even on high-confidence matches.
How fast are credit decisions?
The clean path returns a decision in under a second. Refers go to a human underwriter with the model's reasoning pre-staged so the review is minutes, not hours. Latency and funnel conversion are tracked side by side so a faster decision does not quietly cost approval quality.
What do we own at the end?
Everything. Source code, model artefacts and documentation, deployment infrastructure, runbooks, the audit trail, and the evaluation suites. No vendor lock-in and no ongoing licence fee for what we built. If you ever want to run the system without us, you can.
What is our PCI scope with this system?
We keep it as small as the design allows. Primary account numbers never land in your application tier; we tokenise at the processor edge and keep PCI scope inside the processor's vault. Your audit surface stays narrow and your QSA sees a short list of controls.

AI for Insurtech

Which claims and policy systems do you integrate with?
We build against Guidewire ClaimCenter, Duck Creek Claims, Majesco, and Sapiens through their supported APIs and messaging interfaces, and against custom and legacy policy admin via the integration layer you already govern. Writes land where your change control already lives.
How does this meet NAIC and state DOI expectations on AI?
Every model we ship carries a documented risk tier, a validation pack, a monitoring plan, and a named governance owner, aligned to NAIC's model AI bulletin and the state DOI bulletins that have adopted it. The model-risk register is yours; your committee signs off before production.
Will adjusters trust fast-track recommendations?
Trust is a workflow problem, not a model problem. Every recommendation carries its top contributing signals, its confidence, and a documented reason an adjuster can accept or override. We measure override rate as a product metric and tune to it.
How do you handle fraud screening without over-referring to SIU?
The screening model is tuned to your SIU's historical outcomes, not a vendor benchmark. We measure precision at the top of the queue and work to the referral capacity you actually have, not to a theoretical maximum.
What about catastrophe events, when adjuster capacity is swamped?
Cat protocols are a first-class part of the design. When a cat is declared, triage thresholds tighten, document requests pre-stage, and the fast-track path is governed by the cat rules your claims leadership has approved for that event.
Does this touch rating or binding?
No. The underwriting copilot drafts and surfaces; it does not quote and does not bind. Rating and binding stay with the underwriter and the rate filing your actuaries have made with the DOI.
What do we own at the end?
Everything. Source code, model weights where applicable, deployment infrastructure, the model-risk register, validation evidence, and the evaluation suite. No vendor lock-in and no ongoing licence fee for what we built.
How long until a claims leader feels the difference?
The discovery phase runs two weeks. The prototype phase that follows is designed to put a working, measured system in front of adjusters inside eight weeks from kickoff on a single line of business.

AI for Automotive

Does this touch our ADAS perception models?
We integrate your perception stack into the production flow: scenario coverage, edge-case triage from telemetry, regression on target hardware, and the evidence pack homologation expects. We do not claim to train your perception models from scratch; that is work your ADAS organisation owns.
How does this meet UN R155 and R156?
The system respects the CSMS controls expected under R155 and produces the artefacts your type-approval authority reviews under R156: scoping, canary waves, telemetry verification, and release decision records. Release decisions stay with your release manager.
What about EU telemetry and GDPR?
Purpose, retention, and data-subject rights are enforced in the pipeline for EU VIN cohorts. Data flows out of the EU only on a lawful basis you have already established with your DPO.
Can you integrate with our existing connected-vehicle platform?
Yes. We build to your platform's supported APIs rather than replacing it, and we write back only through the release and campaign tools your R156 process already governs.
Will engineers trust the pattern alerts?
Trust is a workflow problem, not a model problem. Every alert carries its top contributing signals, the affected VIN population, and the prior-model-year analogue where it exists. Detection thresholds are engineer-authored. We measure override rate as a product metric and tune to it.
Does this decide a recall?
No. The system surfaces patterns, drafts the case, and attaches the evidence. Decisions on recall vs service campaign, and the regulator-facing filings, stay with your product safety and legal functions.
What do we own at the end?
Everything. Source code, model weights where applicable, deployment infrastructure, detection thresholds, the evaluation suite, and the evidence packs your authorities expect. No vendor lock-in and no ongoing licence fee for what we built.
How long until a quality engineer feels the difference?
The discovery phase runs two weeks. The prototype phase that follows is designed to put a working, measured system in front of engineers inside eight weeks from kickoff on a single subsystem.

AI for Mobility

Can this run alongside our existing dispatch platform?
Yes. We often start as a decision-support layer that reads from your platform and writes suggestions back, with a dispatcher override on every call. When the measurement proves out, we expand to the surfaces that need to act automatically.
How do you handle driver classification risk?
The system never takes actions that implicitly reclassify a driver. Nudges are advisory, incentives are opt-in, and schedule control stays with the driver. Your policy team configures what is advisory and what is required.
How do you handle price caps and city audits?
Hard caps per city are enforced at the pricing layer; every change publishes a machine-readable reason; a reporting view backs city audits and your own transparency reporting.
What about electric fleets and charging?
Charging is scheduled against forecasted demand troughs where possible, with safety overrides for low state of charge. Charger slots can be reserved when the network predicts a crunch.
What data do you need to start?
A representative sample of trip history, live driver and request feeds in a staging environment, and your current ETA and pricing baselines. We work with what you have; we do not require a clean warehouse to begin.
What do we own at the end?
Everything. Source code, model weights where applicable, deployment infrastructure, documentation, evaluation suites. No vendor lock-in and no ongoing licence fee for what we built.
How quickly can we feel it on the ground?
Discovery runs two weeks. The prototype phase is designed to put a working, measured system against one city, one shift block, within eight weeks from kickoff.

AI for Aerospace

Which maintenance platforms do you integrate with?
We build against the supported interfaces of AMOS, TRAX, and the common MRO platforms. We write back only through surfaces the platform approves and never silently update airworthiness state.
Does this make airworthiness decisions?
No. The system drafts tasks, cites regulations, and queues reviews. Every airworthiness decision is signed by a licensed engineer, and every automated action logs its inputs for audit.
How do you handle ITAR and EAR for defense-adjacent work?
Environments are segregated to cleared personnel, data flows are audited, and model inference runs with no cross-tenant retention. Where required, deployment is on-premise or in a sovereign cloud region.
What sensor and flight-data sources do you work with?
ACARS, QAR, FOQA, and vendor telemetry across common OEM platforms (compatible with GE Digital, Boeing AnalytX, Honeywell Forge). We work with what you have; we do not require a clean sensor stack to begin.
How do you handle false positives?
Confidence thresholds are configured with your reliability leaders; low-confidence predictions are deferred rather than drafted; precision and false-positive rate are measured and reported.
What do we own at the end?
Everything. Source code, model weights where applicable, deployment infrastructure, documentation, evaluation suites. No vendor lock-in and no ongoing licence fee for what we built.
How long until the reliability team feels it?
Discovery runs two weeks. The prototype phase is designed to put a working, measured system against one aircraft type and one component family inside eight weeks from kickoff.

AI for Physical AI

Does this control the line?
No. The system is a quality and routing layer. Safety-rated interlocks remain with the PLC and safety PLC under ISO 13849 and IEC 61508; emergency stops always win.
What about 21 CFR Part 11 and ISO 13485?
In regulated manufacturing, electronic records are signed, immutable, and retrievable by lot and SKU; model versions and operator identity are enforced on every signature.
What happens when a new SKU is introduced?
The perception stack learns the new SKU from a short on-floor capture pass. The station runs in slow mode with a human-pick fallback until a line engineer approves the new model.
Where does inference run?
On an edge compute node at the line, with bounded latency and no reliance on a cloud round-trip for pass or fail decisions. Training and evaluation can run centrally.
How do you handle false rejects?
Per-detection confidence drives routing: borderline calls go to a QA review lane rather than being auto-rejected. Operators can override and the override is recorded with a reason code.
What do we own at the end?
Everything. Source code, model weights, deployment infrastructure, documentation, evaluation suites. No vendor lock-in and no ongoing licence fee for what we built.
How long until the line feels it?
Discovery runs two weeks. The prototype phase is designed to put a working, measured system on one station and one SKU family inside eight weeks from kickoff.

AI for Wellness

Which booking systems do you integrate with?
We build against the booking system you already run: Mindbody, Boulevard, Mangomint, and others, and against custom booking back-ends where you have one. Reads and writes go through the vendor interfaces; we do not bypass the system of record.
Will the assistant give health or clinical advice?
No. The assistant answers questions about services, pricing, timing, policies, and the practitioner roster. Any health, medical, or clinical question is declined and warm-routed to a licensed practitioner with a prepared summary. The assistant also discloses that it is an assistant.
How is GDPR and state privacy handled?
The system operates as a processor under contract. Consent is captured once at the booking surface and respected across channels. Data subject access, correction, and deletion requests are supported. In the US we build against CCPA and the newer state analogues with the same mechanics.
What happens if the member has allergies or sensitivities on file?
Allergies and sensitivities are checked against the service and product list before confirmation. A conflict blocks the confirmation until a practitioner has reviewed it, so a reaction never gets booked out of the system.
Does the system send messages on our behalf without a human review?
No new sequence or bulk message goes out without a named owner reviewing. One-to-one reminders and transactional confirmations run on templates that have already been reviewed. Opt-outs propagate across every channel at once.
Will the assistant sound like our brand or like a vendor template?
It will sound like your brand because it runs on your content, your tone-of-voice guidelines, and your service pages, with evaluation on your actual copy. If it does not read like you, we have not shipped yet.
What do we own at the end?
Everything. The source, the prompts, the retrieval index configuration, the integration code, and the evaluation suite. No vendor lock-in and no ongoing licence fee for what we built.
How long until the front desk feels the difference?
Discovery takes two weeks. The prototype phase that follows is designed to put a working, measured system in front of the front desk and practitioners inside eight weeks from kickoff on a single location.

AI for Travel & Hospitality

Which PMS and channel managers do you integrate with?
We build against the PMS, booking engine, and channel manager you already run: Opera, Mews, Cloudbeds, Apaleo, SiteMinder, and others. Reads and writes go through the vendor interfaces; the channel manager stays the system of truth for distribution.
How is PCI DSS handled?
Card data stays inside the existing PCI DSS boundary of your payment processor. The assistant never sees a PAN. Where a transaction needs to happen in-channel we use tokens or hosted payment links, so the PCI scope is kept tight.
Will the assistant quote rates and give refunds?
The assistant quotes rates only inside the parity and promotional rules the revenue team has approved, and will not override. Comps, refunds, upgrades, and rate overrides stay with a named human. The assistant drafts; a human approves.
How does the dynamic pricing layer avoid breaking parity or contracts?
Rate parity rules, promotional floors, and segment contracts are hard constraints the pricing layer refuses to breach. Moves above a configured delta require revenue-team approval before they publish. Every move is logged with the signals and the reasoning behind it.
What languages are supported, and how is tone handled?
The assistant answers in the guest preferred language where that language is enabled. Tone runs inside your brand style and voice guidelines, with evaluation on your actual content. If it does not read like you, we have not shipped yet.
What about GDPR, CCPA, and local hospitality rules?
The system operates as a processor under contract. Consent for marketing, messaging, and profiling is captured and respected. Data subject access, correction, and deletion requests are supported. Guest-registration, star-rating, and licensing requirements vary by jurisdiction, and we build to the ones your properties operate under.
Can you work alongside our existing RMS, CRS, or chat vendor?
Yes. We often start by replacing one piece (concierge, review response, or pricing proposals) while incumbent systems stay in place, and expand only where the measured outcome justifies it.
What do we own at the end?
Everything. The source, the prompts, the retrieval index configuration, the integration code, the pricing rules, and the evaluation suite. No vendor lock-in and no ongoing licence fee for what we built.

AI for Education

Which SISs and LMSs do you integrate with?
We build against OneRoster for roster, LTI 1.3 for LMS launch and grade-passback, and Clever or ClassLink for identity. That covers PowerSchool, Infinite Campus, Skyward, Canvas, Schoology, Google Classroom, and Microsoft Teams for Education. For older systems we handle direct SQL extracts under a DPA.
Where does student data live?
In your tenant where your SIS already sits, or in a dedicated environment we operate for you under a signed data-processing agreement. Encrypted at rest and in transit, minimum-necessary access, no retention at inference time, and no training on your data.
What about COPPA for under-13 learners?
Any flow that collects data from a learner under 13 runs through verifiable parental consent, with the data categories and retention stated plainly. Consent can be revoked, and revocation purges derived content.
Will teachers actually use this, or revert to doing it themselves?
Adoption is a workflow problem, not a model problem. We co-design with the teachers who will use it: lesson structure, how grades land, where intervention flags surface, how family notes stage. If a teacher cannot sign a draft note in under a minute, we have not shipped yet.
Does the tutor replace the teacher?
No. The tutor supports practice; the teacher owns the mastery decision, the relationship with the learner, and every family contact. We draw that line explicitly in the system, not just in the brochure.
What happens when the model is wrong?
Every grade is a draft until a teacher signs it. Low-confidence mastery calls return 'not yet' rather than a false signal, and every automated action logs the model version and inputs so an audit can reconstruct it. We treat errors as a product metric, not an afterthought.
What do we own at the end?
Everything. Source code, model weights where applicable, deployment infrastructure, documentation, evaluation suites. No vendor lock-in and no ongoing licence fee for what we built.
How long until a teacher feels the difference?
The discovery phase runs two weeks. The prototype phase that follows is designed to put a working, measured system in front of a grade team inside eight weeks from kickoff, on real planning and grading work.

AI for Communication

Which help desks and CCaaS platforms do you integrate with?
We build against Zendesk, Intercom, Salesforce Service Cloud, Genesys, and Five9 as the platforms customers most often already run, using each platform's own APIs and app frameworks. These are compatibilities, not partnerships. For other platforms we use their public APIs.
Where does conversation data live?
In your tenant where your help desk already sits, or in a dedicated environment we operate for you under a signed data-processing agreement. Encrypted at rest and in transit, minimum-necessary access, no retention at inference time, and no training on your data.
How do you handle payment-card data on calls?
PCI DSS flows redact the cardholder number from the transcript and recording at the turn level. The assistant never sees a card in plain text; account actions requiring card data route to your PCI-approved flow.
Will the self-service agent make things worse before it makes them better?
The common failure mode is a bot that answers from training, not from the current policy. Our self-service agent answers from the live knowledge base with citations, and escalates when retrieval does not support a confident answer. We measure deflection-with-positive-CSAT, not deflection alone.
What happens when the model is wrong?
Every reply is a draft until an agent sends it. Low-confidence turns escalate rather than draft, and every automated action logs the model version and inputs so an audit can reconstruct it. We treat errors as a product metric, not an afterthought.
Can you work alongside our existing bot or vendor?
Yes. We often start by replacing one piece (agent-assist, or QA) while an incumbent bot stays in place, and expand only where the measured outcome justifies it.
What do we own at the end?
Everything. Source code, model weights where applicable, deployment infrastructure, documentation, evaluation suites. No vendor lock-in and no ongoing licence fee for what we built.
How long until an agent feels the difference?
The discovery phase runs two weeks. The prototype phase that follows is designed to put a working, measured system in front of a queue of agents inside eight weeks from kickoff on a single channel.

AI for Architecture

Which authoring tools and platforms do you integrate with?
We read from Revit, AutoCAD, Bluebeam, Procore, Autodesk Construction Cloud, and your specification master. We write back only through surfaces each platform approves. The drawing of record stays the drawing of record; the system never edits it directly.
Where do our drawings and specs live?
In your tenant or in a dedicated environment we operate for you, encrypted at rest and in transit. Minimum-necessary access, no retention at inference time, and no training on your project data. The retrieval index enforces a hard project and client boundary.
Will this stamp drawings or decide code compliance?
No. The system drafts replies, summarises specs, assembles proposals, and indexes archives. Every reply, every drawing revision, and every stamped submittal remains the work of a licensed architect. We cite the code; the architect decides compliance.
What happens when the model is wrong?
Every reply is a draft until a project architect signs it. Low-confidence passages are flagged, retrieval refuses when it is not sure, and every automated action logs the model version and inputs so an audit can reconstruct it. We treat errors as a product metric, not an afterthought.
Can you work alongside our existing tools and our existing site?
Yes. We often start by replacing one piece (drawing search, or a portfolio rebuild) while the rest of the stack stays in place, and expand only where the measured outcome justifies it.
What do we own at the end?
Everything. Source code, the retrieval index, deployment infrastructure, documentation, evaluation suites, and the portfolio site. No vendor lock-in and no ongoing licence fee for what we built.
How long until a project architect feels the difference?
The discovery phase runs two weeks. The prototype phase that follows is designed to put a working, measured system in front of one project architect inside eight weeks on a single live project.
Does this replace the project architect?
No. It shortens the search, the drafting, and the proposal assembly so the project architect spends time on the exceptions, on the design review, and on the work only a licensed architect can do.

AI for Real Estate & Construction

Which platforms do you integrate with?
We build against Procore and Autodesk Construction Cloud on the construction side, and Yardi, MRI, and CoStar on the asset side. We honour each platform permissioning and write back only through the surfaces each one approves. The rent roll stays the rent roll; the drawing of record stays the drawing of record.
Where do our drawings, contracts, and camera feeds live?
In your tenant or in a dedicated environment we operate for you, encrypted at rest and in transit. Camera retention follows the project record requirement; the system processes progress against the schedule, not faces. No training on your project data.
Will this approve change orders or sign leases?
No. The system drafts replies, scores impact, routes items for approval, and reports progress. Change orders are routed to the owner for sign-off; leases are signed by the leasing lead; crews are rescheduled by the superintendent. The system informs the decision; it does not make it.
What happens when a camera feed is unclear or interrupted?
The system reports the gap rather than claiming progress it cannot see. Low-confidence reads are flagged, and the superintendent is the final arbiter of what is in place on site.
Can you work alongside our existing scheduling and accounting stack?
Yes. We often start by replacing one piece (RFI triage, or schedule-risk reporting) while the rest of the stack stays in place, and expand only where the measured outcome justifies it.
What do we own at the end?
Everything. Source code, the retrieval index, the vision pipeline, deployment infrastructure, documentation, and evaluation suites. No vendor lock-in and no ongoing licence fee for what we built.
How long until a project executive feels the difference?
The discovery phase runs two weeks. The prototype phase that follows is designed to put a working, measured system in front of one project executive inside eight weeks on a single live project.
Does this replace the superintendent or the project executive?
No. It shortens the search, the drafting, and the reporting so the superintendent and the project executive spend time on the exceptions, on the owner conversation, and on the work only an experienced builder can do.

AI for Retail & E-commerce

Do you work with Shopify, Netsuite, SAP, and the marketplaces?
Yes. We build against the native APIs for Shopify, Netsuite, SAP, Oracle Retail, Klaviyo, and the major marketplaces. We are not a partner of any of them; we are compatible with the stack you already run, and we write back only through surfaces each platform approves.
What catalogue size does the forecasting layer work on?
We have modelled from a few thousand SKUs to catalogues in the hundreds of thousands. The hierarchical approach picks a different model family for fast-movers, slow-movers, and launch SKUs, so a long tail does not drag down accuracy on the hero items.
Will the personalisation copy still sound like our brand?
Brand voice is the ground truth, not an afterthought. All drafts are scored against a versioned brand brief; drafts that fall below threshold are routed back rather than published. The system is only useful to a brand if the brand still sounds like itself a year in.
Do we have to swap out Klaviyo or our CDP?
No. We build to the lifecycle and segmentation tools you already run. The system drafts flows and segments; your lifecycle lead approves and sends. If you want to migrate later, your flows and segments are yours in portable form.
How does this handle marketplace rule changes?
Marketplace rules are enforced in the staging layer. When a rule changes, the affected listings pause for a merchant to re-approve rather than silently publishing something that will get rejected.
What about returns in categories we do not see today?
Return reasons are clustered by SKU and variant, with the three worst offenders surfaced each week. The system drafts the specific sizing, photography, or copy change; the merchant decides whether to ship it.
What do we own at the end?
Everything. Source code, trained models, prompts, forecast pipelines, lifecycle flows, documentation, evaluation suites. No vendor lock-in and no ongoing licence fee for what we built.
How quickly do merchants feel the difference?
Discovery runs two weeks. The prototype phase that follows is designed to put a working, measured system in front of merchants inside eight weeks, on one category and one channel, before any broader rollout.

AI for Media & Advertising

Do you work directly with the platform APIs or through a vendor?
Directly. Meta Marketing API, Google Ads API, TikTok Marketing API, DV360, The Trade Desk, Conversions API, and Measurement Protocol. We are not a partner of any of them; we build you a system that is compatible with the stack you run.
How does this handle iOS tracking and cookie deprecation?
The measurement layer assumes the signal is lower resolution. Server-side event streams run through Conversions API and Measurement Protocol, aggregated attribution is read next to a rolling MMM, and the bridge between platform numbers and MMM is published each week rather than argued about.
What is your position on media-mix modelling?
MMM is back, but on a weekly cycle, not a quarterly one. Platform attribution gives directional signal; MMM gives contribution. Both read in the same brief with a documented bridge, so the team decides against one number instead of two.
Will the creative variants still look like the brand?
The brand brief is the ground truth. Every drafted variant is scored against it; variants below threshold are routed back rather than published. A creative director approves every variant before it ships.
How do you handle brand safety on open exchanges?
Delivery is monitored against your suitability criteria and known-low-quality and MFA lists. The system drafts allow and deny list changes each week; your buyers approve. Nothing is blocked silently, and every change is logged.
Can you work alongside our agency, or do you replace them?
Either. The system is designed to live inside an agency's weekly rhythm as a measurement and creative velocity layer, or inside an in-house team the same way. We often start by owning one piece (the Monday brief, or the creative pipeline) while the rest of the team keeps running.
What do we own at the end?
Everything. Source code, trained models, MMM pipelines, prompts, documentation, evaluation suites. No vendor lock-in and no ongoing licence fee for what we built.
How quickly do performance leads feel the difference?
Discovery runs two weeks. The prototype phase that follows is designed to put a working measurement bridge and creative pipeline in front of the team inside eight weeks, on one brand or client, before any wider rollout.

AI for Marketing

What does owned outright actually mean?
At handover, the repo is yours, the deploys run on infrastructure you control, the data lives in your tenant, and the runbooks are in your hands. There is no licence fee for what we built, no gatekeeper, and no change order required to make a change. We can keep supporting you on fixed terms afterwards, but you are never dependent on us.
Why do you build platforms instead of running a retainer?
Retainers treat launch as the finish line and the client as a tenant. We build the opposite: platforms you own after launch, not ones you keep renting back from the people who built them. Our incentive is to leave you with something that works without us.
Do you work with Webflow, WordPress, Contentful, HubSpot, Klaviyo, Marketo?
Yes. We build to the platforms you run and are compatible with the stack you already have. Where you want to migrate off a vendor to an owned platform, we do that too; your content and your flows are portable at handover.
What is AEO and GEO, and why do they matter now?
AEO is answer-engine optimisation: structured so an engine can return a direct answer with your brand attached. GEO is generative-engine optimisation: structured so an LLM or generative search surfaces and cites you. Buyers are already asking in both; the brands that invested in SEO ten years ago are the ones that will invest in AEO and GEO this year.
Will the AI-assisted content still sound like our brand?
Brand voice is the ground truth. Every draft is scored against a versioned brand voice brief; drafts below threshold are routed back rather than published. An editor approves every piece. The pipeline is only useful if the brand still sounds like itself a year in.
How do you resolve attribution between the CRM and the dashboards?
Pipeline-sourced revenue is reconciled against the CRM. Multi-touch, self-reported, and MMM read next to each other with a documented bridge. Finance signs off on the bridge each quarter; the brief flags divergences rather than hiding them.
What do we own at the end?
Everything. Source code, trained models, prompts, content pipelines, lifecycle flows, attribution reconciliations, documentation, evaluation suites. No vendor lock-in and no ongoing licence fee for what we built.
How quickly does the marketing team feel the difference?
Discovery runs two weeks. The prototype phase that follows is designed to put a working piece in front of the team inside eight weeks, on one pillar or one lifecycle flow, before any wider rollout. The platform ownership itself lands at handover and does not go away afterwards.
Objections answered? Let's talk.