Forward Deployed Engineer - GenAI

ABOUT:Builds generative AI applications — LLM-powered features, RAG pipelines, and enterprise search that ship to production, not just a demo.KEY RESPONSIBILITIESBuild GenAI applications — LLM-powered features, copilot/chat experiences, enterprise searchDesign and implement RAG pipelines: chunking strategy, embedding selection, hybrid retrieval, re-ranking, GraphRAG where structured retrieval is neededFine-tune and adapt models (LoRA/QLoRA) when prompt engineering and RAG aren't sufficientEngineer and version production prompts; build prompt/context management into the application layerIntegrate LLM APIs (OpenAI, Anthropic, Azure OpenAI) and open-source model endpoints with auth, rate-limiting, and cost controlsInstrument applications for evaluation — output logging, quality scoring, human-feedback loopsOptimize latency and token cost through caching, batching, and model routing strategiesTranslate client business requirements into concrete GenAI feature specificationsCommunicate technical tradeoffs (cost, latency, accuracy) to non-technical product stakeholdersCollaborate with the Agentic AI Architect and Data Scientists on shared componentsDocument architecture and prompt design decisions for handoff and maintainabilityREQUIREMENTS & SKILLS4–8 yrs software engineering, with 1–3 yrs hands-on GenAI/LLM application buildingStrong Python; experience with LangChain, LlamaIndex, or equivalent orchestration frameworksVector databases and embedding strategies (Pinecone, Weaviate, pgvector), plus knowledge-graph/graph-database tooling (Neo4j) where relevantUnderstands LLM failure modes (hallucination, context-window limits, cost blowup) and designs mitigationsExperience with model fine-tuning techniques (LoRA/QLoRA) and evaluation harnessesHands-on with enterprise GenAI/agentic platforms — Microsoft Azure AI Foundry, AWS Bedrock (incl. Strands Agents SDK), and Google Vertex AI; open-source frameworks (LangChain, LlamaIndex) a good-to-have where no platform is mandatedAPI design and integration experience, including auth, rate limiting, and streaming responsesFamiliarity with prompt-versioning and LLMOps tooling (LangSmith, Weights & Biases, or similar)Clear technical writing — documents a RAG architecture for a non-technical stakeholderComfortable working directly with client engineers during embedded deliveryCollaborative — works with architects, data scientists, and QA without needing everything pre-specifiedAdaptable under ambiguity — prompt-based systems require rapid iteration and tolerance for imperfect first attempts