Custom AI Agent for Knowledge & Support
A conversational agent that answers from a company's own documents and acts on its internal APIs — running entirely on-premise so no data leaves the building.
The challenge
Teams wanted an AI assistant over their internal knowledge — procedures, tickets, product docs — but sending that content to a hosted LLM API was a non-starter on both privacy and cost. Cloud per-token pricing also made high-volume, always-on use hard to budget.
Approach
- Local models — served with Ollama on the client’s own hardware; a domain-specific model fine-tuned with QLoRA where accuracy needed a lift.
- Retrieval — documents chunked and embedded into a ChromaDB vector store, so answers are grounded in the client’s real content with citations.
- Orchestration — a FastAPI service ties retrieval, the model and the client’s internal REST APIs together, so the agent can look things up and take actions.
- Interface — a lightweight React chat UI, embeddable in existing tools.
Result
Staff get answers in seconds instead of digging through wikis or waiting on a colleague; the whole stack runs inside the client’s network with no external API dependency and no usage-based cloud bill.
Add the deployment context (which client, timeline) and real usage numbers from the admin panel.
Want an outcome like this?
Instant answers for staff and customers, no per-token cloud bill, and full data privacy from local model hosting.
Book a callMAIE — medical availability integration engine