StartPilot

Problem

Give companies and startups a single support hub — onboarding, documentation, operational support, automations — without depending on external LLM providers or exposing sensitive data to third-party services.

Architecture

• Svelte + SvelteKit frontend for both the customer dashboard and the operator workspace: chat, tickets, knowledge base, automations.
• FastAPI Python backend: REST APIs, Pydantic validation, LLM orchestration, conversation and knowledge-base management.
• PostgreSQL as source of truth: companies, users, tickets, conversations, versioned knowledge base.
• vLLM as the self-hosted inference engine: high-throughput serving of open-weight LLMs, dynamic batching, KV-cache management.
• RAG pipeline on PostgreSQL + pgvector to fetch context from the knowledge base before inference.

Stack

Svelte · SvelteKit · TypeScript · FastAPI · Python · PostgreSQL · vLLM

Outcome

On-brand support served on the customer’s own infrastructure, with no data leaving for third parties. Typical response latencies under 2 seconds thanks to vLLM batching, knowledge base editable by the operator with immediate effect on the answers.