MCP Got My Agents Connected. Here's Everything I Had to Build Around It
MCP made it easy to describe a tool once and let the model call it; making those tool calls hold up in production was a separate job, and that is where most of my time went.
Senior engineer across backend, cloud and front end, with AI where it helps.
CalcMenu Philippines · 2021–2023
I led the rebuild of CalcMenu Cloud, alone at first and later with a team of four developers. The legacy system was slow and buggy, ran outdated and vulnerable software, and couldn’t integrate with other systems.
The new platform is made of microservices organised by business domain behind an API gateway, with RabbitMQ queues for long-running tasks, shipped with CI/CD and Docker on Azure and AWS.
Old and new systems ran in parallel: core features first, then read-only data migrations that clients verified, then clients moved over wave by wave. Service was never interrupted.
Most of that work has been full-stack: Python, ASP.NET and TypeScript services, Angular and React front ends, on Azure and AWS. I led the CalcMenu Cloud rebuild, first alone and then with a team of four. Today I work on Nooko.ai, an AI culinary platform, at EGS.
I’m based in Manila and work remotely with teams across Asia and Europe, overlapping with European hours. I speak English and Turkish.
— Salih
Field notes
Practical write-ups from building and running production software.
MCP made it easy to describe a tool once and let the model call it; making those tool calls hold up in production was a separate job, and that is where most of my time went.
Practical lessons from deploying AI agents with tool-calling. Hallucinated arguments, infinite loops, runaway costs, and the guardrails that actually prevent them.
Production-focused guide to building RAG pipelines with Python. Chunking strategies, vector search with pgvector, re-ranking, prompt construction, and the stuff that breaks past the tutorial stage.