Run a 27B model on your laptop: Bonsai 2, locally
Prism ML’s ternary-quantized Bonsai 2 squeezes a 27.8B model into about 6 GB. The exact local setup — which quant to pick, the llama.cpp fork you need, and the commands that actually run it.
Writing
No listicles. These are problems I hit in real Flutter and AI work — how we found them, what we changed, and what I would do differently next time.
Featured essay · Flutter
Most Flutter architectures are diagrams. This is the version I ship: feature-first boundaries, typed failures, and tests that protect the seams.
Inside the essay
Prism ML’s ternary-quantized Bonsai 2 squeezes a 27.8B model into about 6 GB. The exact local setup — which quant to pick, the llama.cpp fork you need, and the commands that actually run it.
Alibaba opened Qwen4-Max-Preview on Nov 11 — the first Qwen4 model, aimed at coding and agentic work, with a reported 1M-token context. What is confirmed, what is only reported, and how to evaluate it.
On Nov 12–13, Outlook, Teams, Word and Excel went down worldwide — servers, not a cyberattack. If your product leans on one cloud, here is the resilience checklist I keep coming back to.
Jensen Huang resigned as Nvidia CEO on Nov 12, with Colette Kress interim. The confirmed timeline, what is still unexplained, and how to think about vendor risk if your stack runs on CUDA.
One canonical profile, content as data, prerendered metadata, and audits that gate a deploy — the boring structure that keeps a portfolio true as it grows.
Chunking by structure, hybrid retrieval, reranking, validated citations, and the operations work that keeps an AI answer trustworthy after launch.
Six patterns that survived real apps, from AsyncNotifier screen state to route-scoped providers, plus the habits that quietly make Riverpod codebases hard to change.
Latency budgets, streaming, citations, prompt versioning, and failure states — the unglamorous work that makes an AI feature safe to ship.
A practical process for measuring cold start, cutting the main() budget, and avoiding the optimizations that do not survive a release build.
What building a Gujarati-first app with five languages taught me about scripts, RTL, plurals, content operations, and store listings.
I am not a Google Developer Expert yet. This is the deliberate 12-month plan I am running — and the parts of public work that actually count.
Release trains, a device matrix, and a written decision log — how mobile delivery changed shape when I stopped being the senior developer and became the lead.
Work with me