About the roleThere's a wide gap between an agent that works in a demo and one that works across millions of live transactions. Closing it is the job. You'll embed with the teams and customers who depend on AI - risk, fraud, collections, payments, support, developer experience - and design, build, and ship agentic systems into their production environments. You'll also help build the platform underneath: Paytm's AI inference platform (Pi) and the agentic runtime, orchestration, and tooling that lets agents reason, plan, use tools, and run multi-step workflows safely.
What you'll do- Embed & deploy
- Tackle greenfield problems alongside internal teams and customers - scope ambiguous needs and build agents from scratch that fit how they actually work.
- Own deployments end-to-end: discovery, build, integration, activation, and the tuning that earns trust and adoption.
- Lead pilots and demos, drive adoption, and clear blockers before they stall a rollout.
- Build agentic systems
- Architect agentic systems - reasoning, planning, tool use, memory, multi-agent coordination - that run real workflows with guardrails.
- Build safe tool-use infrastructure across APIs, databases, and services, with permissioning, sandboxing, and human-in-the-loop.
- Ship SDKs, patterns, and reusable blueprints so internal teams build and deploy agents fast.
- Make it reliable
- Design and run rigorous evals: measure quality, catch regressions, and feed results back into the system.
- Build observability, tracing, and guardrails that prove agents are safe and keep them safe as models and data drift.
- Own the multi-model inference your agents depend on (text, voice, code, vision) - latency, throughput, and cost.
- Lead
- Set technical direction and standards for agentic systems; mentor engineers and partner with ML, product, and security.
What you'll bring- 5+ years in software engineering, with 3+ in AI systems or LLM applications, and production systems shipped end-to-end.
- Strong grasp of LLM agent architectures (ReAct, RAG, tool use, multi-agent) and hands-on agentic orchestration and evaluation.
- Proficiency in Python across a broad stack - pipeline, agent, service, and instrumentation.
- Production experience on AWS and Azure with containerized deployments (Docker, Kubernetes).
- Strong customer and stakeholder instincts; able to impose structure on ambiguity and push back when needed.
- A bias toward shipping and comfort operating without a clean spec.
- Solid understanding of agentic security risks (prompt injection, privilege escalation, data leakage).
- Strong written and verbal communication.
Nice to have- Agentic systems in regulated industries (fintech, payments, credit, healthcare).
- Cloud AI/ML services (AWS SageMaker / Bedrock, Azure ML / Azure OpenAI); multi-cloud or hybrid.
- MCP or agent communication standards; agent evaluation and observability tooling.
- Model serving (vLLM, TensorRT-LLM, Triton), fine-tuning, quantization, or LoRA.
- Workflow orchestration (Temporal, Airflow, Prefect) for AI workloads; voice / multimodal / edge inference.
- Testing and verification for non-deterministic AI systems.
Be among the first to define how agentic AI ships across a company running payments and credit at massive scale - with direct line of sight from your work to the outcome, and broad ownership across both the platform and the field.
Go Big or Go Home!
