Blog

Writing

Competition diaries, technical deep dives and things I have learned along the way. Mostly about GNNs, generative AI and developer tooling.

Internship · 27 Jul 2026 · 7 min read

On-Device RAG with Foundry Local: My Microsoft AI Innovators Project

The technical follow-up to April's acceptance post: how I built an offline RAG assistant with Foundry Local, SQLite and Streamlit. Five-layer architecture, chunking decisions, the two distinct failure modes of a 1.5B on-device model, and what my 26-question evaluation set actually measures.

Read the post →

More posts · 4

Eleven Agents for Public-Sector Paperwork: Why We Chose Orchestration Over a Single LLM Call

A technical walkthrough of the multi-agent document-processing system we built for the TEKNOFEST 2026 AI Language Agents competition: why we skipped agent frameworks, what the orchestrator's three conditional gates do, why retrieval reports absolute similarity, and where an adversarial test set broke the system.

VARIANT-GNN: TEKNOFEST 2026 AI in Health — Technical Deep Dive

I wrote about the architecture of the hybrid GNN + ensemble system we developed as team XYRA3 for TEKNOFEST 2026 AI in Health, the design decisions, and the competition process.

What Are Graph Neural Networks (GNNs) and When Should You Use Them?

I explained GNNs from scratch: how graphs represent data, the message passing mechanism, and why they outperform classical models in genomic data.

I Built a Portfolio Site with Claude Code

I wrote about how I developed this site using Claude Code, the tool's strengths and weaknesses, and what I learned in the Next.js + Tailwind combination.