Senior Product Engineer, Agent Systems

This is product engineering on Aedeon, the agent platform Mactores uses to deliver enterprise AWS modernisation, rather than client delivery work. The posting frames the problem sharply: agents that generate plausible code from priors are easy, while agents that modernise a regulated enterprise's estate and can prove the new behaviour matches the old are the hard problem. That framing shapes the architecture. You would build and ship specialised agents in the fleet, meaning parsers, business rule extractors, dependency mappers, test synthesizers and behaviour replayers, plus the orchestration that wires them together. The design constraint is that agents work against a code intelligence graph rather than generating from priors, so every action an agent takes traces back to a specific line of code. You would also implement what Mactores calls governed autonomy, running agents through shadow, supervised and autonomous stages with human review at each, and orchestrate frontier models from Anthropic, OpenAI and Google through the platform's decision model.

What the posting requires

Core Python and systems

- Strong Python with 4+ years in production: asyncio, performance optimisation and idiomatic code
- A solid grasp of distributed systems concepts: state machines, retries, idempotency and eventual consistency
- Integrating language model APIs from Anthropic, OpenAI or similar in production code, covering streaming, function calling, structured output, retries and prompt management
- Comfort with FastAPI or an equivalent async web framework

Cloud and infrastructure

- Working knowledge of AWS: EKS, ECS Fargate, S3, DynamoDB, Lambda, Secrets Manager and CloudWatch
- Docker and Kubernetes, including writing Dockerfiles, Helm charts and manifests
- CI/CD pipelines, with GitHub Actions preferred
- Comfort navigating multi account AWS environments across dev, UAT and production

Testing and automation

- Test driven development as a discipline, with tests written before the code
- Hands on pytest, integration testing and end to end testing
- The ability to design behaviour verification harnesses: dual run, output comparison and equivalence proof
- Using AI tools such as Claude, Copilot or LLM based test generators to improve test quality

Explicitly not required

The posting states that you do not need prior Amazon Bedrock or LangChain experience. What it wants is a strong Python engineer who will learn agent frameworks properly. That makes this genuinely open to backend engineers who have not worked on agents before, which is unusual for this kind of role.

What you would actually be doing

- Building and shipping specialised agents: parsers, business rule extractors, dependency mappers, test synthesizers, behaviour replayers and the orchestration between them
- Designing agents that operate against a code intelligence graph so every action traces to a line of code
- Implementing governed autonomy through shadow, supervised and autonomous stages with human review at each
- Orchestrating frontier models from Anthropic, OpenAI and Google, choosing the right model per task
- Owning delivery of your agents from prototype through deployment and post release validation

The interview process, as published

Mactores lists its process openly, which is rare and worth preparing for specifically. There is a hands on technical session with a panel member covering your skills, experience and how you deliver quality under pressure, then a structured panel interview covering experience, product thinking and how you work, then a 30 minute HR discussion about the offer and next steps. The posting also asks applicants to answer as many of the application questions as possible to speed things up.

Location and working pattern

Mumbai.

A good fit if

You are a strong Python engineer curious about agent systems but have been put off by postings that demand specific framework experience.

Think twice if

You dislike heavy test discipline. Testing runs through this entire posting, including test driven development as a stated requirement and the ability to build equivalence proof harnesses.