I build the parts of AI systems that don't show up in the demo the routing logic, the memory boundaries, the eligibility gates that decide whether a request should even reach a model in the first place.

Most of my recent work has been on Bob, a multi-agent platform automating financial risk and compliance workflows for enterprise teams. It taught me that the hard part of agentic AI is rarely the model itself it's the deterministic scaffolding you build around it: state management, structured handoffs between agents, and keeping five independently-reasoning components aligned on what the user actually asked for.

Currently: writing up real production problems and the trade-offs behind how I solved them in production-agentic-ai not tutorials, actual case studies. A payload-size bug that silently broke SSO. Why free-text handoffs between agents drift over time. Why context budget matters more than context size.

Stack I reach for: Python, LangGraph, FastAPI, Azure AI Search / OpenAI embeddings, OpenTelemetry for tracing pipelines that fail silently instead of loudly.

Background: a few years shipping backend and AI systems (Zinnia, 9NEXUS), BE in Computer Engineering. Based in India, open to relocating for the right role in the UK or EU.

If you're working on similar problems multi-agent consistency, RAG at enterprise scale, making probabilistic systems debuggable I'd genuinely like to compare notes.

·