Programming and foundations

•

- Strong Python. Practical working use of at least one of TypeScript / Java / Go.

•
- Solid SQL and data modelling; comfortable with both relational and vector stores.

•
- Sound software engineering fundamentals — testing, version control, CI/CD, code review discipline.

GenAI core (must be hands-on, not conceptual)

•

- LLM application development — prompt design and prompt engineering as an engineering discipline,

structured output, context management, token/cost optimisation.

•

- RAG — chunking and indexing strategy, hybrid and semantic search, re-ranking, query rewriting, grounding and citation, retrieval evaluation. Awareness of when RAG is the wrong answer.

•
- Agentic systems — tool use, planning and decomposition, multi-agent orchestration, state and memory management, error recovery and retries, MCP or equivalent tool-integration standards.

•
- Evaluation and LLMOps — building eval harnesses, LLM-as-judge with its caveats, tracing and observability

(Langfuse, LangSmith, Arize or equivalent), regression testing on prompt and model changes, monitoring in production.

•

- Model landscape — practical judgement across frontier and open models; multi-model routing; understanding of the cost/quality/latency trade-off rather than brand loyalty.

•
- LLM safety — prompt injection and jailbreak mitigation, data exfiltration risk in tool-using agents, hallucination mitigation patterns, guardrails and validation layers.

Frameworks and tooling

•
- LLM orchestration: LangGraph / LangChain / LlamaIndex / Semantic Kernel or equivalent — and the judgement to know when a framework is unnecessary overhead.

•
- Vector / search: pgvector, FAISS, Pinecone, Weaviate, Azure AI Search, OpenSearch or similar.

•
- Cloud AI platforms: at least one of AWS Bedrock / Azure AI Foundry / Google Vertex AI at production depth.

•
- Containerisation and deployment: Docker, Kubernetes basics, serverless patterns.

•
- Data and pipelines: Pandas,



Airflow / Databricks / equivalent workflow orchestration.

•
- AI-assisted development tooling (Claude Code, Cursor, Copilot) used seriously as a productivity multiplier, not as a novelty.

4.

Evidence We Look

For

This matters more to us than the keyword list above. Strong candidates will be able to walk us through:

•

- A GenAI system they personally shipped to production — its architecture, what broke, and what they changed as a result.

•
- A concrete number: accuracy or quality improvement, cost per transaction reduced, latency brought down,

manual effort eliminated.

•

- An evaluation strategy they designed — how they knew the system was actually working, and how they caught regressions.

•
- A time they argued against using an LLM for something, and what they recommended instead.

•
- Something they built for reuse that other teams actually adopted.
- Good to Have

•

Experience in a client-facing consulting, professional services, or Forward Deployed Engineer model.

•

- Classical ML background — model lifecycle, feature engineering, forecasting — as context, not as the core of the role.

•
- Contributions to open source, technical writing, conference speaking, or an active community presence in the

AI space.

•

- Relevant certifications (cloud AI, Anthropic, or equivalent), treated as supporting evidence rather than a substitute for shipped work.

6.

Experience and Education

- 6+ years of relevant technology experience overall (typically 6–14, but we will not screen out robust candidates on either side of that).
- 2+ years hands-on with LLM-based systems in production. We are deliberately not asking for more.

Production LLM application development is roughly three years old as a discipline — anyone claiming a decade of it is describing something else. Depth and evidence here outweigh total years.

•

- Demonstrated experience leading technical teams and mentoring engineers.

•
- BE / B.Tech / MCA / M.Tech, or equivalent demonstrated capability. We will interview strong self-taught engineers

📌 Gen AI Lead (Pune)
🏢 R Systems
📍 Pune