Job Summary
This engineer integrates an LLM (Claude API) into two parts of the pipeline: an offline classification pass that labels component intent during reference board ingestion, and an online fallback reasoning pass that generates placement recommendations when no reference pattern is found. Output from the LLM feeds directly into downstream automated logic, so reliability, structured output, and failure handling are critical.
Key Responsibilities
Required Skills:
• Claude API or equivalent (Anthropic SDK / OpenAI SDK), with experience generating structured, schema-constrained output (JSON) from LLM calls
• Prompt engineering for classification tasks: given component type, value, net names, and placement context, classify intent and signal class reliably
• Validation and failure handling: detecting malformed output, hallucinations, and low-confidence responses; defining fallback behavior
• RAG pipeline development: vector embeddings, similarity search, ChromaDB or pgvector
• Python
Skill Requirements
Strongly Preferred:
• Production experience building LLM pipelines where model output drives automated downstream actions (not just conversational interfaces)
• Retrieval quality tuning: chunking strategy, embedding model selection, hybrid search
• AWS Bedrock or similar managed LLM deployment
• Experience with confidence scoring and explainability — the system shows LLM reasoning to the designer, so output must be human-readable and trustworthy