Staff AI Platform Engineer - Inference & Agentic Systems

About the Role

We are a small team of AI builders in Paytm Labs.

As a Staff AI Platform Engineer, you will work across inference and agentic systems. You will

contribute to Paytm's AI inference platform (Pi), serving internal teams and enterprise customers

- running our own coding and domain-specific models (voice, vision, risk, fintech workflows) as

well as third-party models. You will also architect and build the platform that enables

autonomous AI agents to operate safely and reliably in production - the runtime, orchestration,

and developer tooling for agents to reason, plan, use tools, and execute complex multi-step

workflows, automating both software development and business processes.

You will work at the intersection of LLMs, distributed systems, and production fintech

infrastructure, helping define how inference and agentic AI are built and deployed across

payments, risk, fraud, collections, support, and developer experience.

What You'll Do

• Inference & Model Serving
• Build and operate multi-model serving across modalities (text, voice, code, vision) on shared infrastructure
• Own the model lifecycle: download, deploy, serve, monitor, update, swap
• Drive inference optimization: latency, throughput, cost - including quantization, batching, caching, and routing strategies
• Ensure inference is fast and reliable for the agents and systems that depend on it
• Agentic Systems
• Architect and build the Agentic AI Platform - runtime infrastructure, orchestration systems, and developer tooling for autonomous agents
• Design multi-agent coordination systems enabling agents to collaborate and solve complex workflows
• Build robust tool-use infrastructure that allows agents to interact with APIs, databases, and services safely
• Implement workflow automation: agents that execute multi-step business and engineering tasks with appropriate guardrails
• Build safety and guardrail systems including permissioning, sandboxing, and human-in-the-loop workflows
• Develop evaluation and observability frameworks to measure agent behaviour, detect regressions, and debug failures
• Develop SDKs and APIs that allow internal teams to build and deploy agents quickly and safely
• Platform & Technical Leadership
• Define technical direction and architecture for agentic systems across the organization
• Build patterns and standards for agent design, tool calling, and evaluation
• Partner closely with ML, product, and security teams to deliver production-grade agent systems
• Mentor engineers and contribute to best practices for agent system design

What You'll Bring

• 8+ years of software engineering experience, with 3+ years in AI systems or LLM applications
• Strong understanding of LLM-based agent architectures: tool use, multi-step workflows, multi-agent coordination, and their failure modes
• Experience building highly reliable distributed systems
• Experience evaluating LLM systems in production: building evals, detecting regressions, and debugging non-deterministic failures
• Proficiency in TypeScript or Python, and willingness to work in both: the agent platform is TypeScript on Bun with Temporal workflows on Kubernetes and EC2, the inference platform is Python.
• Experience working with modern LLM APIs or open-source models
• Experience with or strong interest in model serving (vLLM, TensorRT-LLM, Triton)
• Understanding of distributed systems: task queues, event-driven architectures, state management, and durable long-running workflows
• Experience with cloud platforms (AWS, GCP) and containerized deployments
• Strong understanding of security risks in agentic systems (prompt injection, privilege escalation, data leakage)
• Demonstrated experience leading complex technical initiatives
• Strong written and verbal communication skills

Nice to Have

• Experience building agentic systems in regulated industries (fintech, healthcare, enterprise)
• Familiarity with Model Context Protocol (MCP) or agent communication standards
• Experience with model fine-tuning, quantization, or LoRA
• Experience building CI/CD automation and developer tooling
• Experience adapting workflow orchestration systems (Temporal, Airflow, Prefect) for AI workloads
• Experience with voice models, multimodal models, or edge inference
• Experience designing human-in-the-loop or oversight systems