Why a blog
The portfolio shows what I shipped. The blog shows how I think while shipping it.
I work on production AI systems — multi-agent pipelines, fine-tuned LLMs, low-latency inference serving 500K+ transactions a month. Most of what I learn doesn't fit on a one-line résumé bullet:
- Why a Phi-3 LoRA at 4-bit beat a 70B baseline for one specific compliance task
- How the 97% accuracy number was actually 84% before three rounds of negative-mining
- Why we ripped out a vector DB and replaced it with a 2 MB JSON file
Those are the posts. Real numbers, real trade-offs, real bugs — not LinkedIn-flavored "I'm excited to announce."
What you'll find here
Three buckets:
- War stories — production failures, debug logs, the fix nobody documents
- Pattern notes — multi-agent orchestration, tool calling, eval harnesses
- Tooling reviews — what I picked, what I dropped, why
Posts are short by default. If a topic doesn't fit in 800 words, it probably wasn't ready.
How this is built
This blog is MDX — Markdown with embedded React. Everything is statically generated at build time, so each post is its own indexable page. Source on GitHub.
// the entire pipeline, more or less
const posts = readdirSync('content/blog')
.filter(f => f.endsWith('.mdx'))
.map(parseFrontmatter);
That's it. No CMS, no DB, no admin panel. The posts live in a folder and Git is the only history I need.
Subscribe?
There's no newsletter yet. Follow @Shrijeeth on GitHub or check the LinkedIn — I'll cross-post the meaty ones.
Otherwise: bookmark /blog, come back when a title catches your eye.
The Matrix is a system, Neo. That system is our enemy. — Morpheus
But also: most systems are fine if you measure them.