AI Engineer - Agentic & GenAI Systems

Role Summary Design, build, and operate production-grade agenticand GenAI systems—end to end. You’ll ship services (not just notebooks): robustAPIs, reusable components, and secure pipelines that connect LLMs, tools,knowledge, and enterprise systems. You’ll pair strong software engineering withmodern AI practices (RAG, agent orchestration, policy chains, evals) to delivermeasurable business outcomes at scale.

Responsibilities

Agent & Application EngineeringS Design multi-agent systems (MAS) with planning, tool-use, and delegation (e. g. , LangGraph/Semantic Kernel); expose them via REST/gRPC APIs (FastAPI/Express/Java/Go).

Implement tool adapters (SQL, search, document stores, web calls, code exec) with strict type contracts and safe sandboxes. Build model gateway integrations (OpenAI/Azure OpenAI/Bedrock/Vertex; self-hosted vLLM/TGI) with routing, rate-limits, retries, and fallback chains.

Orchestrated workflows with LangChain or LlamaIndex — planning, tool calling, structured extraction, multi-step state Typed contracts on every model output using Pydantic schemas, so downstream code never parses free text Guardrails and policy enforcement at the boundary, with deterministic rules taking precedence over the model Retrieval, Data & Knowledge Stand up RAG services: chunking, enrichment, embeddings, indexing, hybrid/vector search (pgvector/Pinecone/Weaviate; OpenSearch/Azure AI Search).

Implement ingestion pipelines (Airflow/Prefect/Celery/Ray) for docs, tickets, chat, and ERP/CRM data; handle PII redaction and metadata governance. Optimize retrieval quality (chunking strategies, re-rankers, query rewriting) with offline/online evaluation and A/B tests.

Quality, Testing & Evaluation Treat prompts and graphs as code: version, diff, and test them (unit tests for prompts/tools; golden sets; regression suites). Build evaluation harnesses (latency, cost, accuracy, toxicity, hallucination, guardrail hit-rates); wire into CI. Add drift detection for conversational systems; im