Writing
Back to homepage insightsWriting that explains the engineering judgment behind the shipped work.
These entries keep the portfolio from stopping at screenshots and summaries. They show how Umar reasons through model behavior, orchestration, retrieval, and product tradeoffs in plain language.
Featured
The strongest reading paths stay visible first.
Featured entries are the best starting points for understanding the thinking behind Umar's applied AI work.
Feb 12, 2026
My First Attempt at Fine-Tuning an SLM using Unsloth
A hands-on breakdown of turning a small language model into a focused information extraction system for NER-style tasks.
Feb 5, 2026
What I Learned Building a Transformer Model from Scratch
A practical reflection on implementing transformer mechanics directly to demystify how the architecture actually works.
Jan 27, 2026
I Built an AI Crew to Automate My Job Applications. Here's What I Learned.
A build log on orchestrating AI agents to automate repetitive application work while keeping human judgment in the loop.
Archive
The rest of the writing archive remains easy to scan.
Posts can stay external-first on Medium while still being modeled locally and routed consistently inside the portfolio.
Jan 18, 2026
Beyond Keywords: How I Taught an AI to Recommend Books Based on Vibe
An applied recommendation-system experiment focused on translating subjective intent into more useful retrieval signals.
Jan 5, 2026
Chat With Your Financial Data: Building a SaaS Metrics Copilot with a Local LLM
A walkthrough of building a local-LLM copilot that turns financial questions into structured analysis over business data.
Jul 9, 2025
Building My First MCP Server: Connecting Claude to PostgreSQL
A practical article about using Model Context Protocol to connect an LLM workflow to structured PostgreSQL data.