AI solutions
AI assistants that are safe to put in front of customers
Five assistants I have built for WhatsApp, Instagram, Messenger and web chat. Each recording is the real software running; demos use fictional businesses and are labelled as such.
The model talks, the code decides
Prices, availability, stock and orders come from typed tools over a deterministic backend. The model never computes money.
Grounded or silent
Assistants answer only from the business's own data and say so when they cannot. Citations where it matters.
A human one message away
Handoff is a first-class feature: pause the agent, flag the thread, let staff reply, hand back.
Measured, not hoped for
Evaluation sets and adversarial runs before release, so behaviour changes are visible.
AI Receptionist
A 24/7 WhatsApp booking agent for appointment-based businesses.
- Books against real availability. Checks open slots, prevents double bookings with a database exclusion constraint, and confirms in chat.
- Reminders that act. 24-hour and 2-hour reminders with buttons to reschedule or cancel without leaving WhatsApp.
- Guardrails. Refuses medical advice, admits when it does not know, never invents a policy or a price.
- Human handoff. "Can I speak to a person?" pauses the agent, flags the thread in the inbox, and lets staff reply then hand back.
- TypeScript
- NestJS 11
- Next.js 15
- Prisma
- PostgreSQL 16
- Redis + BullMQ
AI Shopify Sales Agent
Selling and fulfilment inside WhatsApp, Instagram and Messenger.
- Sells from the real catalogue. Searches live Shopify products by category, colour, size and price, and checks stock per variant before promising anything.
- No order without a yes. Order creation requires the full summary and an explicit confirmation. Totals are recomputed in code at that moment.
- COD into Shopify. Creates the order in the merchant's admin with delivery charge and payment method recorded, tagged by channel.
- Courier booked automatically. Books the parcel, stores the tracking number in Shopify, and messages the customer from dispatch to delivery.
- TypeScript
- NestJS
- Next.js
- PostgreSQL
- Redis + BullMQ
- Claude API
AI Real Estate Agent
Qualifies buyers on WhatsApp and books viewings that are genuinely free.
- Qualifies before it books. Captures budget, area, size and timeline, and keeps what the customer said separate from what the agent inferred.
- Only verified listings. The model cannot invent a property or a price; it can only reference search results.
- Slots that are really free. Agent hours, calendar blocks and travel time between viewings all feed the availability engine.
- Fair-housing safe (US). Matching uses objective criteria only. Neighbourhood, mortgage and lease questions go to a person.
- TypeScript
- NestJS
- Next.js
- PostgreSQL
- Redis + BullMQ
- Claude API
AI Ordering Agent
WhatsApp ordering for restaurant chains where the backend owns the money.
- Understands real orders. Per-item drinks and sizes, multiple corrections in one sentence, Roman Urdu. Options resolve on the server.
- Backend owns the money. Cart, tax, delivery fee and promotions are computed in code; the model relays the numbers.
- Branch routing. A typed address or a shared location pin goes to the nearest open branch with the right fee and ETA.
- Rule-based upsell. One relevant add-on per conversation, with acceptance tracked.
- TypeScript
- Node.js
- SQLite / PostgreSQL
- Claude API
- WhatsApp Cloud API
Tailyr AI Concierge
A grounded, multi-store RAG assistant for high-value retail.
- Strictly grounded. Two confidence gates and per-source citations on every answer.
- Knowledge loop. Unanswered questions go to a partner dashboard; answers are embedded back and attributed to the team member.
- Store isolation. Multiple brands on one platform; a store can never see another store's data.
- Sold-stock intelligence. Knows what was previously sold for comparisons but never offers it as available.
- TypeScript
- NestJS 11
- Next.js 15
- Supabase (PostgreSQL + pgvector)
- OpenAI embeddings and chat models
- Railway
Have a process customers already do over chat?
Bookings, orders, enquiries, support. Send me a description and I will tell you what an assistant could safely take over, what it should not, and what the first working prototype would look like.
Discuss an AI assistant