About the Role
We’re looking for an experienced AI Engineer who is passionate about building intelligent, tool-using agents leveraging cutting-edge LLM orchestration frameworks. You will be a core contributor to the design, development, and deployment of agentic systems that combine reasoning, memory, retrieval, and orchestration to power the next generation of enterprise AI applications.
Core Technical Responsibilities
Agentic Design
- Proven ability to architect multi-step agents with planning, tool usage, and self-reflection.
- Experience with orchestration frameworks like LangGraph, CrewAI, or custom graph/task-based planners.
LangChain Ecosystem
- Deep understanding of Chain, Runnable, and LCEL abstractions.
- Built custom output parsers, retrievers, and memory components.
- Experience with tracing & evaluation tools such as LangSmith, Langfuse, or equivalents.
Agent Memory
- Implemented short-term memory (e.g., message history, token windowing).
- Designed long-term memory using vector stores (FAISS, Pinecone, Milvus, OpenSearch) or knowledge graphs.
- Strong grasp of performance/cost trade-offs across memory backends.
Retrieval-Augmented Generation (RAG)
- Designed hybrid retrieval pipelines (BM25 + dense vector search).
- Familiar with chunking/embedding strategies, adaptive retrieval, and reranking.
Prompting & Function Design
- Skilled in structured prompting (JSON/YAML schemas, function-calling).
- Experienced in dynamic prompt injection for real-time tool use.
LLM Evaluation & Safety
- Knowledge of automated evaluation techniques (rubric-based, self-critique).
- Familiar with red-teaming, toxicity, bias detection, and PII leakage prevention.
Programming & Tooling
- Proficient in Python 3.11+ (typing, Pydantic, asyncio) and TypeScript.
- Test-driven development with pytest, coverage tools, and LangChain-specific testing patterns.
AWS Infrastructure (Strong Preference)
- Hands-on with AWS Bedrock, SageMaker JumpStart, and HuggingFace DLCs.
- Designed serverless workflows using Lambda, Step Functions, and EventBridge.
- Experience deploying to ECS/Fargate, EKS with GPU acceleration.
- Uses CloudFormation/CDK or Terraform with best-practice IAM and cost monitoring.
MLOps & Observability
- Experience in vector-store migrations, prompt/model versioning, and model registries.
- Monitoring with OpenTelemetry, CloudWatch, Honeycomb, or similar observability stacks.