GTI Doctor — AI Assistant

🤖 The GTI Doctor AI assistant is now LIVE. Ask it questions about geotechnical instrumentation — sensor specifications, installation procedures, application checklists — and it answers using grounded citations from the 18 reference manuals in this knowledge base (5,032 indexed chunks, 12.5 MB of text).

Visit: https://gti-doctor.henry-phamduc.workers.dev/

Open full chat ↗ Try chat inline ↓

Example questions

• What is a vibrating-wire piezometer and how is it installed in a borehole?
• Give me the slope-stability instrumentation checklist from the FHWA manual.
• How does the wireless mesh-network backhaul work?
• Which instrument would I use to monitor lateral movement in a 30 m deep excavated retaining wall?

How it works

Component	Technology
LLM	@cf/meta/llama-3.1-8b-instruct-fp8 (Workers AI, free tier)
Embeddings	@cf/baai/bge-m3 (1024-dim multilingual)
Vector database	Cloudflare Vectorize (gti-doctor-embeddings index, 5,032 vectors)
Worker runtime	Cloudflare Workers (with Workers AI + Vectorize bindings)
Grounding	RAG pipeline — embed query → top-8 chunks → context-enriched prompt

The assistant is trained only on this knowledge base. If a question is outside scope, it will say so instead of inventing an answer.

API

Endpoint	Method	Purpose
/api/health	GET	Service heartbeat
/api/info	GET	Worker configuration
/api/chat	POST	RAG chat (streaming via SSE or batch JSON)
/api/search	POST	Pure semantic search (no LLM, just top-K chunks)

Example:

curl -X POST https://gti-doctor.henry-phamduc.workers.dev/api/chat \
  -H "Content-Type: application/json" \
  -d '{"question":"What is a vibrating-wire piezometer?","stream":false}'

Status

Component	State
Corpus ingestion (18 PDFs, 5,032 chunks)	✅ Complete
Vectorize index created	✅ gti-doctor-embeddings (1024-dim, cosine)
Cloudflare Worker deployed	✅ https://gti-doctor.henry-phamduc.workers.dev/
Chat endpoint	✅ Verified — 200 OK, grounded answers with citations
Chat widget embedded in intranet	✅ Live iframe above