Skip to content
Sadiq Khan

Writing

18 essays on backend systems, AI engineering and the occasional side quest. Most started on LinkedIn.

2026

  1. Oct 5 · After hours · 1 minI made a simulated fruit fly try to walk24,115 real neurons from the first complete fly nerve cord wiring diagram, wired to a 3D body in a physics engine, on a laptop.
  2. Sep 7 · After hours · 1 minMy first PyCon, in HiroshimaThree days at PyCon JP 2026, a Django booth, and the strange feeling of finally meeting a community I had only known through code.
  3. Aug 12 · After hours · 1 minSome weeks, walk away from the buildThree days somewhere signal is a rumour, and the questions that get quieter there.
  4. Aug 4 · After hours · 1 minFive stickers for the Django boothDesigning a small collectible sticker series for the Django Software Foundation booth at PyCon JP 2026.
  5. Jul 28 · Work · 2 minYour eval dashboard is measuring nothingPublic benchmarks are saturated. The three measurements teams shipping agents actually rely on, and what each one costs you.
  6. Jul 23 · Work · 2 minIdempotency keys are not enoughThe Idempotency-Key header protects an API boundary, not the workflow underneath. Three ways durable execution engines solve the rest.
  7. Jul 20 · Work · 2 minParallel agents don't share a prompt cacheFan out five agents at t=0 and you pay for five cache writes. How Anthropic, OpenAI and Gemini cache differently, and the one move that works for all three.
  8. Jul 2 · Work · 1 minFrom chunks to atomsAgent teams are quietly moving off chunks plus top-k. What they need looks less like retrieval and more like structured state with provenance.
  9. Jun 24 · After hours · 2 minWhen the verifier is cheaper than the generatorGoogle's hidden quantum optimisation, a public contest that grades circuits without revealing the answer, and why any verifiable problem is now on a clock.
  10. Jun 4 · After hours · 1 minPokémon in your editorA weekend side project, a VS Code extension, and what small joyful tools teach you about distribution.
  11. May 19 · Work · 2 minThe oldest algorithms in your RAG stackLouvain, PageRank and a 50-token trick: how much classical computer science is quietly powering 2026 retrieval.
  12. May 12 · Work · 1 minYou might be optimising the wrong layerA paper reached the top of SWE-bench with 12% fewer tokens by evolving the harness, not the model or the prompt.
  13. May 5 · Work · 1 minContext isn't retrieved. It's assembled.Similarity is not understanding. Why AI integrations that shine in demos fall over in production.
  14. Apr 29 · After hours · 2 minCareCoord: a multi-agent care coordinatorWhat I built as a finalist at Anthropic's Built with Opus 4.7 hackathon: parallel specialist agents that read a family's medical papers and connect what no single doctor sees.
  15. Apr 27 · Work · 2 minSix multi-agent patterns nobody tells you aboutBuilding with AI agents isn't prompting. It's distributed systems design where one of your components happens to think.
  16. Jan 30 · Work · 1 minOutages are loud. Data corruption is quiet.Moving 10M+ CRM records across distributed systems, and why the failure mode worth designing for isn't downtime.

2025

  1. Nov 23 · Work · 1 minREFRAG and the cost of long contextMeta's REFRAG compresses retrieved chunks into embeddings and keeps only the high-value tokens raw. Why that matters for RAG in production.
  2. Oct 16 · Work · 1 minOne million requests, six backend stacksStress-testing Rust, Go, Node.js, Erlang and Elixir at a million concurrent requests, and what runtime design does to performance.

Contact

sadiqkhan795@gmail.com

Say hello. I read everything.