AI Engineer — Agentic Systems
Build the AI agents at the heart of Codly — LLM-powered agents that plan, call tools and act on AWS, Azure, GCP and Kubernetes, under policy and with a human approving what matters.
Read the full job description
About the role
You'll design, build and ship the agents that run our customers' clouds: operations, FinOps, compliance, service management and OS/DB agents coordinated by an AI supervisor and a policy engine. This is production agentic AI for banks and insurers — reliability, safety, cost and auditability matter as much as model quality.
What you'll do
- Design and build multi-agent workflows — planning, tool and function calling, memory, retries and human-in-the-loop approval steps.
- Build tool integrations that let agents act safely on cloud APIs (AWS, Azure, GCP), Kubernetes, databases and ITSM systems.
- Build retrieval (RAG) over runbooks, policies and compliance frameworks using embeddings and vector databases.
- Create evaluation harnesses — offline test sets, regression suites, guardrails and hallucination checks — and track accuracy, latency and cost per task.
- Route work across models (OpenAI, Azure OpenAI, Gemini and open models), and optimise prompts, context and token spend.
- Ship production services in Python with tests, observability and tracing; own features from design to customer rollout.
- Work with security on prompt-injection defence, data isolation and audit logging for every agent action.
What you'll bring
- 3-6 years of software engineering, including 1+ year building LLM-powered applications that real users depend on.
- Strong Python (async, APIs, testing) and solid software design fundamentals.
- Hands-on with LLM APIs and an agent framework such as LangGraph, LangChain, LlamaIndex, Semantic Kernel or AutoGen — and the judgement to know when not to use one.
- Practical experience with RAG, embeddings and vector databases (Milvus, pgvector, Pinecone or similar).
- Working knowledge of at least one public cloud (AWS, Azure or GCP), Docker and Kubernetes basics.
- SQL (MySQL or PostgreSQL) and caching (Redis).
Nice to have
- Cloud certification (AWS, Azure or GCP).
- Experience with Model Context Protocol (MCP), fine-tuning or serving small open models.
- Exposure to FinOps, security or compliance domains (RBI, SEBI, ISO 27001, PCI DSS).
- Open-source contributions or public writing about LLM systems.