Generative AI Engineer

• 1. Strong Python + FastAPI — async endpoints, background tasks, streaming responses, pydantic models. This is the backbone of every app.

• 2. LLM/agent engineering — practical experience with agent frameworks (pydantic-ai, LangGraph/LangChain, or similar), prompt design, tool-calling, and RAG. Critically: understanding failure modes — hallucination, context contamination, non-deterministic output. Much of the recent work has been debugging exactly these.

• 3. RAG in depth — chunking, embeddings, vector search (Qdrant/pgvector/etc.), retrieval quality tuning, and citation/grounding. Knowing why retrieval returns the wrong thing.

• 4. React (hooks) + solid CSS — building and, importantly, fixing real UI: layout, responsive behavior, streaming state. Not just wiring components — being able to look at a screen and know why it's broken.

• 5. Concurrency & state discipline — a recurring class of bugs here has been shared mutable state / callback bleed under concurrent requests. Someone who instinctively reaches for request-scoped state, locks, and thread-safety.

• 6. Docker + Kubernetes fluency — comfortable with containers, deployments, debugging pods/logs. Doesn't need to be a platform expert but must operate the dev loop.

• 7. Testing rigor — writes pytest + vitest, uses mocks/fixtures, tests before shipping. This codebase has real test coverage and expects it maintained.

• 8. Debugging & reading code — the single most valuable trait. Able to trace a symptom ("answers the wrong question") to a root cause across frontend, backend, and agent layers.

Nice-to-haves

• AWS Bedrock / model gateway experience specifically (vs. only OpenAI).

• Langfuse or LLM observability tooling.

• python-pptx / document generation (PowerPoint, PDF) — a niche but real part of this codebase.

• Regulated/pharma or life-sciences domain familiarity (oncology, diagnostics, regulatory terms like FDA PMA/510(k), IVDR) — speeds up prompt and content work enormously.

• Helm chart authoring, GitHub Actions/CI.

• Design sensibility — the project cares about premium, non-generic UI; someone with taste (not just CSS mechanics) is a force multiplier.

• Vector DB ops — Qdrant collection management, embedding pipelines, SharePoint/data-source sync.

• Prompt-eval / LLM-testing frameworks for measuring answer quality over time.