What You Must Have Actually Done

Not just what you know. What you have shipped.

- Deployed 2–3 agent-based systems in production - stateful, multi-step, real users
- Used LangGraph for multi-agent orchestration with memory, tool routing, and state management
- Built projects where AI (Claude Code, Codex, Cursor) wrote significant portions of the code
- Implemented RAG pipelines end-to-end - chunking, embedding, retrieval, re-ranking, evaluation
- Integrated agents with real enterprise APIs - not just OpenAI playground or sample data
- Debugged a production agent failure - and fixed it without blaming the model
- Can articulate when NOT to use agents - that is how we know you have built things

Bonus - Real Differentiators

- Experience with Claude Code CLI in team environments (CLAUDE.md, shared context, multi-session flows)
- Familiarity with LangSmith for agent tracing, evaluation pipelines, and debugging at scale
- Has shipped something using MCP (Model Context Protocol) or similar shared-context tooling
- QA/testing mindset for agents - systematic evaluation of non-deterministic outputs
- Background in IT services or consulting - managing client expectations while building
- Experience with SLMs, fine-tuning, or on-device/edge agent deployment

What We Are Not Looking For

- Someone who lists LLMs on a resume but has only called the API in a Jupyter notebook
- AI enthusiasts whose hands-on experience is less than a year old
- People who explain everything in terms of frameworks they have never deployed
- Consultants who can only narrate what others have built

Key Responsibilities

Delivery & Architecture





- Own end-to-end delivery of AI-native programs - from architecture through production deployment
- Design and build multi-agent orchestration systems using LangChain, LangGraph, CrewAI, or equivalent
- Integrate agent systems with enterprise surfaces: APIs, ERPs, CRMs, data platforms - not toy datasets
- Define agent topology: tool routing, memory strategy, state machines, fallback handling

Agentic Coding & Development

- Run agentic coding workflows using Claude Code, Cursor, OpenAI Codex, or equivalent CLI tools
- Lead projects where AI writes significant portions of the codebase - and you guide, review, and ship it
- Work with CLAUDE.md, shared context frameworks, and multi-session agent setups for team use
- Debug non-deterministic agent outputs systematically - not by gut feel

Client & Stakeholder Engagement

- Translate business problems into agent architectures for global CXO-level stakeholders
- Run discovery workshops, solution reviews, and delivery cadences with client teams
- Prepare and present technical proposals, POC plans, and roadmaps - own the story end-to-end

Team & Practice

- Mentor junior AI engineers; raise AI engineering quality across the delivery team
- Stay current: evaluate current models, frameworks,



and tooling before the hype catches up
- Contribute to internal knowledge bases, reusable frameworks, and accelerators

Skills

Agent Orchestration LangChain, LangGraph, CrewAI - not just conceptual Agentic Coding Tools Claude Code CLI, Cursor, OpenAI Codex, Copilot RAG &

• Vector Stores Chroma, Weaviate, Pinecone - knows where RAG breaks LLM APIs &

• SDKs Anthropic, OpenAI, Gemini - prompt design, tool use Python / TypeScript Primary languages for agent + backend development LangSmith / Observability Tracing, evaluation, debugging agent runs Cloud Platforms Azure, AWS, GCP (at least one) - deployment, infra, managed services API &

• System Integration REST, gRPC, Kafka - enterprise integration patterns MCP / Shared Context Model Context Protocol, CLAUDE.md, Beads Agent Evaluation Testing non-deterministic outputs, guardrails, evals CI/CD &

• DevOps Git, containers, pipelines - agents need to ship Client Communication Can present architecture to a CXO without jargon How We Will Evaluate You

Not a theory round.

Expect to walk through something you have actually built - architecture decisions, what broke in production, what you would do differently. If you cannot do that with specifics, this role is not the right fit.

Evaluation stages

- Stage 1 - Technical screen: Walk us through a live agent system you built
- Stage 2 - Architecture discussion: Given a business problem, design an agent solution on the spot
- Stage 3 - Stakeholder simulation: Present your approach to a non-technical executive audience

📌 Full-Stack Gen AI Engineer (Pune)
🏢 Zensar
📍 Pune