Senior, hands-on AI engineer to design, build, and productionize GenAI applications end-to-end. You'll lead the
development of robust LangChain/LangGraph agentic workflows, high-quality RAG pipelines, and scalable
microservices on Google Vertex AI. You'll own system design, implementation, MLOps, observability, and governance
—partnering closely with product, data, security, and platform teams to deliver reliable, secure, and cost-efficient AI
products.
Key Responsibilities
Architecture & Orchestration
Design multi-step agentic workflows with LangGraph (state machines, tools, retries, timeouts) and
LangChain (chains, tools, memory).
Build guardrails (input/output filtering, red-teaming hooks) and observability (tracing, telemetry,
logging, prompt/version tracking).
● RAG Pipelines
○ Own ingestion pipelines: chunking, embeddings, document normalization, metadata, and vector DB
indexing (e.g., Pinecone, Weaviate, Milvus, FAISS).
○ Implement retrieval strategies: hybrid (BM25 + dense), multi-vector, reranking, query planning,
LangGraph retrieval sub-graphs, caching.
○ Build domain-specific adapters (schema, ontology alignment) and grounding with structured
tools/knowledge bases.
● Vertex AI & Platform Engineering
○ Productionize services on Google Vertex AI (Models, Endpoints, Workbench, Pipelines, Vector
Search, Feature Store).
○ Containerize with Docker, orchestrate with Kubernetes/GKE, and automate with CI/CD (GitHub
Actions/Cloud Build).
● Full-Stack Delivery
○ Build user-facing apps (React/Next.js) and backends (Python/FastAPI, Node/Express), including
authentication/authorization and rate limiting.
○ Develop tooling/services (e.g., document loaders, evaluators, red-teaming flows, prompt
versioning, synthetic data pipelines).
● Evaluation & Reliability
○ Define and automate GenAI evaluation: relevance, faithfulness, hallucination rate,
answer-exactness, latency, cost.
○ Use techniques like RAGAS, G-Eval, rubric-based human-in-the-loop, pairwise comparisons, A/B
tests, and production feedback loops.
● Security, Governance & Cost
○ Implement data privacy controls (PII detection, masking), policy enforcement, prompt hardening,
and audit logging.
○ Optimize latency and TCO (embedding/model selection, batching, caching, streaming, adaptive
routing, quantization where applicable).
● Mentorship & Standards
○ Establish best practices for prompt patterns, orchestration, testing (unit & scenario), and model
lifecycle management.
○ Mentor engineers; collaborate with product/design to scope features and deliver business impact.
Required Qualifications
● 7-10+ years software engineering experience; 3-5+ years applied ML/GenAI building production systems.
● Expert with LangChain and LangGraph (tools, agents, state graphs, retries, sub-graphs, observability).
● Hands-on with Vertex AI (Foundational models, Endpoints, Pipelines, Vector Search, Model Garden; IAM &
service architectures).
● Strong RAG practitioner (chunking strategies, embeddings, hybrid retrieval, rerankers like Cohere/Rerank or
bge-rerank, evaluation).
● Deep experience with vector databases (Pinecone, Weaviate, Milvus, FAISS) and embedding models
(OpenAI, Vertex, Cohere, bge-large).
● Production backends in Python (FastAPI) or Node.js, plus React/Next.js front-end experience.
● Solid cloud experience (GCP preferred; AWS/Azure a plus), Docker/Kubernetes, and CI/CD.
● Strong understanding of GenAI evaluation (RAGAS, G-Eval, rubric scoring), observability
(LangSmith/Llamaindex observability/OpenTelemetry), and prompt/version management.
● Knowledge of security & governance: PII handling, isolation, data residency, prompt injection defenses,
secret management.
● Excellent communication; proven track record turning ambiguous problem statements into shipped products.
Nice to Have
● Knowledge graphs (RDF/OWL), retrieval planning, and toolformer/agent patterns.
● LLM serving and routing (DG/mixture-of-experts, function/tool calling, Guardrails, Instructor schemas,
Pydantic).
● Llamaindex experience; structured RAG (SQL/Graph RAG); function/tool calling integrations (Databases,
SaaS).
● On-prem/vector-optimized deployments; GPU utilization, quantization, LORA fine-tuning.
● Experiment tracking (Weights & Biases), feature stores, offline/online evaluation pipelines.
● Enterprise integrations (SharePoint, Confluence, Salesforce) and document governance
