Hackathon Scope
What we shipped in this build, what we intentionally postponed, and why those tradeoffs were the right call.
We treated scope like a feature, not an afterthought. If we tried to build every enterprise checkbox during hackathon time, we would have delivered a beautiful pile of unfinished ambition.
Included in This Build
- Resource ingestion pipeline (markdown, text, PDF, URL, transcript, conversation JSON)
- Chunking + embedding + vector storage in Actian Vector DB
- `search_docs` tool integration for agent workflows
- `/ingest` and `/query` API endpoints
- Memory index (`conversation_memory`) for session-aware retrieval
- Demo UI + API + DCLI + MCP integration surface
- Observability score/alert endpoints for retrieval quality and hallucination tracking
- Configurable alert thresholds and notification hooks
Deferred On Purpose
- Full OCR for scanned/image-only PDFs
- Authentication and full multi-tenancy
- Real-time streaming responses
- Native Confluence and Notion connectors
- Grafana/Prometheus production dashboards
- Cloud deployment hardening and remote control-plane migration
Decision Notes
Why retrieval quality came first
For this project, quality of answers is the product. Shipping auth before grounding would be like painting a race car that has no engine.
Why observability made the cut
It is easy to demo one good answer. It is hard to keep quality stable over time. Observability gives us signal instead of vibes.
Why deployment hardening is deferred
Production hardening matters, but doing it too early would slow the core loop. We built a strong retrieval foundation first so scale work later is worth it.