jobs in the wild.

The problem is not prompt quality alone. The hard part is building the system around the model: High-quality context pipelines Retrieval and ranking over large codebases Eval harnesses for accuracy and regressions Tool orchestration and agent runtime reliability Observability, latency, and cost controls Codebase knowledge graphs that capture real architectural context A big part of this work is building Swiggy’s codebase knowledge graph so agents can reason better about services, APIs, data flows, dependencies, and implementation patterns. Good fit if you have experience in: Backend or distributed systems AI coding agents or developer tools Retrieval, context engineering Evals, benchmarking, and harness design Static analysis, code intelligence, or graph-based systems Building platforms that are simple, reliable, and production-grade