job-hunt · Case Study
job-hunt Case Study
A deep dive into job-hunt: three cooperating subsystems — a LangGraph evaluation graph, a discovery service, and a Playwright application assistant
Pipeline Run Simulation
Run log from evaluation to application
> Waiting to start simulation..._
Fit Score
AI Technology, Applied
Shipping cutting-edge AI in real projects, not just concepts
Prompt Engineering
System-prompt design, structured outputs, graceful fallback. In job-hunt, every LangGraph node uses prompt engineering to return structured JSON, with retries and fallbacks when the model truncates.
RAG Architecture
The full advanced RAG flow — hybrid retrieval (BM25 + dense, fused with RRF), cross-encoder reranking, Neo4j graph multi-hop, and fail-closed grounding — built in the LearnArken project so answers link back to their source passages verbatim and refuse rather than fabricate.
Agent Architecture
A stateful LangGraph agent: ~20 typed nodes, conditional branching, and persisted state. The job-hunt evaluation graph is the practice — distinct from a one-shot linear chain.
MCP Protocol
Model Context Protocol implementation. This website includes an MCP server allowing AI assistants to query portfolio data via standard protocol.
Vertex AI Integration
Powering this website's AI chat assistant with Google Cloud Vertex AI (Gemini). Implemented streaming responses, system prompt injection, and context management.