Writing
Ideas, experiments and lessons from building AI systems — by Farooq Khan and the team. Published from Pakistan, read globally.
How I built a production RAG pipeline in 48 hours
The architecture decisions, trade-offs, and gotchas from shipping a retrieval-augmented generation system for a real client in under two days — and what I'd do differently.
How I built a production RAG pipeline in 48 hours
The architecture decisions, trade-offs, and gotchas from shipping a retrieval-augmented generation system for a real client.
AI in Pakistan: what's really happening
A ground-level view of the AI ecosystem — the talent, the gaps, and what needs to change for Pakistan to compete on the global AI stage.
5 open source tools every AI engineer should know in 2026
A curated list of libraries that have quietly become essential in production ML workflows — with honest notes on each.
How we helped a fintech team cut inference costs by 60%
A walk-through of the model quantisation, caching strategies, and architectural changes that halved their monthly GPU bill.
Fine-tuning LLMs on a budget: what actually works
LoRA, QLoRA, and prompt engineering compared honestly — with benchmarks from real experiments, not toy datasets.
Building a real-time AI chatbot with Next.js and streaming APIs
A step-by-step guide to wiring up a streaming LLM response into a polished web UI — including edge cases most tutorials skip.
Case study: AI-powered document processing for a legal firm
How we built a secure, on-premise document understanding system that processes 5,000+ contracts a month with 94% accuracy.
Why your AI project failed (and it's not the model)
After reviewing dozens of failed AI initiatives, the same patterns emerge. Almost none of them are technical problems.