Responsibilities

Design and implement state-of-the-art ML and LLM-powered features for the Aera Platform.
Build and own agentic workflows end to end — multi-step reasoning, tool use, subagent orchestration, and long-horizon autonomous loops — with human-in-the-loop checkpoints where the stakes demand them.
Build the eval harness before you build the agent. Own offline and online evaluation, golden datasets, regression gates tied to prompt and model versions, and test suites that hold up under non-determinism.
Optimize agent performance and cost across the full set of levers: context engineering and compression, prompt caching, parallel and async tool calls, model selection and routing, structured outputs, and token budgets.
Instrument what you ship. Build the tracing and observability that lets anyone profile agent behaviour, find the bottleneck, and prove a regression — rather than argue about it.
Design tool and skill surfaces that models can actually use correctly, including MCP servers and reusable agent skills.
Operationalize data science: integrate models into robust pipelines, inference paths, and serverless infrastructure that survive real enterprise load.
Treat agent security as part of the design, not a review step — tool permissioning, sandboxing, and prompt-injection resistance.
Collaborate closely with Data Science, Engineering, and DevOps to deliver solutions others can maintain.
Explore and integrate emerging AI techniques, and bring back a point of view on what's real and what's hype.

About You

B.E./B.Tech in Computer Science, Computer Engineering, or a related field.
3–5 years in software engineering and architecture.
At least 2 years designing and deploying ML or LLM-based systems, including 6–12 months on LLM-specific work.
You architect and own complex, high-stakes systems that orchestrate multiple components — and you can point to the design docs and architectural decisions you wrote to get there. You've owned something from architecture through production reliability, not just to launch.
You are a systems thinker. You look across business domains, find the problem that's actually being solved several times over, and abstract it into building blocks the whole platform can stand on. You step one click out from the problem in front of you and interrogate the assumption underneath it.
You are a power user of agentic coding tools — Claude Code, or equivalent agent harnesses — with real intuition for where models are strong, where they fail, and how to tell the difference before it reaches production. You bring engineering discipline to agent-generated work: you review it, you gate it, you are accountable for it. We care that you've hit the failure modes, not that you've installed the CLI.
You are fluent in current agentic engineering practice, not last year's. Context engineering, tool and skill design, subagent patterns, agent memory, evals and LLM-as-judge, structured outputs, prompt caching, RAG as one retrieval technique among several.
Strong Python. FastAPI or equivalent for production services.
Experience with large datasets, ML pipelines, and distributed systems (Ray, Spark, or equivalent).
Hands-on with PyTorch, Hugging Face, scikit-learn, pandas.
Containerized microservices (Docker, Kubernetes) and CI/CD (Git, Jenkins, Jira).
Humble and adaptable about code and frameworks. LangGraph or comparable orchestration frameworks are useful; none of them are the skill.
Excellent problem-solving, communication, and collaboration.

Good to Have

GoLang for high-performance components.
Vector databases (Opensearch, Pinecone, Weaviate, FAISS, pgvector).
Event streaming and caching (Kafka, Pulsar, Redis).
Durable Execution platform like Temporal
Agent observability and experiment tracking (Langfuse, LangSmith, OpenTelemetry, MLflow, W&B, DVC).
Fine-tuning where it genuinely beats prompting and context — and the judgment to know when it doesn't.
Multi-modal AI: text, image, and structured data in one workflow.
Serverless AI infrastructure on AWS, GCP, or Azure.
“We need more systems thinkers, people who can look across all the business domains and abstract that to, here's the building blocks we're going to need.”

How We Work

Enterprise decision intelligence is unforgiving. When our platform recommends an action, a real supply chain moves. That constraint shapes how we hire and how we build.
Talent density over headcount. We would rather solve a hard problem with a small team of people who are excellent at what they do and better at working together, than staff around the gap. This is the non-negotiable — everything else here depends on it.
Craft is still scarce, still decisive. AI has made it easy to produce code. It has not made great engineering common. There's a difference between writing lines of Python and understanding how code, systems, and products actually work — and the second one is not going away. We hire for the second one.
Context, not process. When something goes wrong, our instinct is a blameless retrospective and a smarter person, not a new approval step. We expect you to take real risks, recover fast when they don't land, and argue for the best outcome for the business rather than the safest one for you.
Comfortable in the discomfort. We are rebuilding how enterprises make decisions while the underlying technology changes under us every quarter. If ambiguity energizes you rather than stalls you, you'll do the best work of your career here.
AI fluency at every level. Not a mandate handed down — an expectation we hold for ourselves too, including for people who no longer write code.

0 - 0 INR a year