HeidelAI
Role
Founding Engineer
Timeline
Jun 2024 — Jun 2025
Status
Live
Stack
Next.js · Python · FastAPI
Dealers and distributors lose deals in the gap between an enquiry arriving and someone getting to it. HeidelAI closes that gap: the agent answers on whichever channel the customer used, works out what they actually want, and replies from the business's own catalogue and pricing rather than a generic script.
What I owned
I was the founding engineer and the product started as an empty repository. I architected and delivered the full stack — schema, services and interface — then led a cross-functional engineering team through the whole product lifecycle to a launch with paying customers.
The messaging layer
Every channel lands in one thread. That meant owning the WhatsApp and Instagram API integrations end to end — webhook ingestion, delivery receipts, media handling and the rate limits that come with the Meta platform — behind a WebSocket inbox that keeps a conversation live no matter where the customer started it.
Multi-tenancy and billing
Every business on the platform gets its own catalogue, pricing and conversation history, and none of it may leak. I designed the tenant isolation model and the subscription payment infrastructure underneath it, so onboarding a customer is a row rather than a deployment.
Retrieval, not guesswork
Replies and product recommendations are grounded in the tenant's own data through a LangChain and Pinecone retrieval stack. The difference shows up in the numbers: response times down 45%, manual handling effort down 60%, and a 20–25% lift in conversion.