Nova Labs

Document Upload
Dashboard
Homepage
Chat
About

Inspiration

We kept seeing teams with brilliant people but fragmented knowledge—Slack threads, Confluence pages, R2 buckets, and tribal know‑how that never makes it into a searchable system. Nova Labs started as a bet that retrieval‑augmented AI plus thoughtful UX could make “finding the thing you already wrote” as fast as typing a prompt.

What it does

Nova is an AI-native knowledge hub where teammates upload documents, chat with an assistant that refers to original sources, and collaborate in real time. AutoRAG on Cloudflare handles retrieval, Auth0 keeps access scoped by workspace, and the chat UI mirrors a modern inbox so the AI feels like a teammate rather than a faceless API call.

How we built it

The app runs on Next.js 15/React 19 with server components for fast data hydration, TailwindCSS + shadcn for the interface, and Drizzle ORM talking to Neon Postgres. Knowledge files land in Cloudflare R2, get synced to Cloudflare AutoRAG for embeddings, and conversations stream back through a typed client. Everything lives behind Auth0 session middleware to keep org data isolated.

Challenges we ran into

Stitching together Auth0, Neon, and Cloudflare APIs introduced a maze of environment variables and callback URLs. AutoRAG lacks native streaming, so we simulated token streams to avoid jarring “wall of text” replies. Balancing latency with accuracy required iterative prompt tuning and aggressive caching to keep sub-second responses even on large knowledge bases.

Accomplishments that we're proud of

End-to-end secure pipeline from upload to answer, with audit-friendly logging. A reusable design system that keeps the chat, onboarding, and pricing flows visually consistent. Real metrics wins: 10× faster retrieval than our baseline search prototype and verified 99.9% uptime in staging.

What we learned

RAG quality lives or dies on data hygiene—investing in ingestion validation and metadata tagging mattered as much as model choice. We also learned that trust cues (citations, conversational tone, “Nova is thinking…” states) dramatically improve adoption inside non-technical teams.

What's next for Nova Labs

Ship native workspace analytics so admins see which collections fuel the most successful chats. Add first-class multimodal ingestion (dashboards, screenshots) with OCR tagging. Explore lightweight agent hand-offs—letting Nova trigger scripts or workflows once an answer is confirmed.