PiSolved Chat
A solo, end-to-end RAG chatbot with a ChatGPT-like experience over domain knowledge.
PiSolved Chat is a private client engagement. The live platform is the proof; implementation details are available to walk through under NDA.
Users needed domain-specific answers with the fluency of ChatGPT, but grounded in a specific knowledge base - with persistent threads and sensible usage limits.
An end-to-end RAG chatbot I built entirely solo: real-time streaming chat, thread-based persistent sessions, daily quotas, and an email workflow for quota upgrades.
Real-time streaming interface, thread-based persistent sessions
Context-aware responses via the OpenAI Assistants API
Full schema in PostgreSQL with Drizzle ORM
Daily quota management + Resend email workflow for upgrade requests
Streaming chat
Real-time token streaming with auto-scroll and loading states.
Persistent threads
Thread-based sessions that survive reloads.
Quota system
Daily quota management with a Resend-powered upgrade flow.
Grounded answers
Domain-specific responses via the Assistants API.
| Metric | Value |
|---|---|
| solo build | 0-7 wk |
| end to end, solo | 0% |
| experience at the time | Under 1yr |
| grade delivery | Prod |
I built this entirely solo in 6-7 weeks: the streaming chat UX, thread persistence, the OpenAI Assistants integration, the full PostgreSQL/Drizzle schema, the daily quota system, and the Resend email workflow. This was the project that confirmed my transition from frontend to genuinely full-stack.
- RAG quality is mostly retrieval and grounding discipline, not model choice.
- Quotas and upgrade flows are part of the product, not an afterthought.
Want the deeper story?
Ask my AI about PiSolved Chat for specifics, or reach out directly.
