001 Product-minded AI engineer · systems builder
AI engineer, product builder, and technical writer focused on automation-heavy tools, production AI systems, and software that solves real operational friction.
002 Featured
A product for LinkedIn hiring workflows that collapses setup friction and accelerates applicant analysis — install, land authenticated, evaluate at scale. Recruiting is the domain; the work is systems design.
An internal sourcing and evaluation engine spanning multiple data sources — ingestion pipelines, document extraction, AI evaluation at scale, and post-evaluation enrichment.
A production RAG and document-intelligence system: async streaming backend, multi-format parsing, structured chunking, and vector retrieval tuned for real-time Q&A.
003 Capabilities
Tools where AI does real work — scoring, parsing, evaluation, retrieval — inside a product people actually use. Front to back: the model, the pipeline, and the interface around it.
Automation-heavy tooling that compresses multi-step operations into fast, reliable flows.
RAG backends with careful parsing, chunking, and retrieval quality — built for production, not demos.
Structured evaluation at scale, plus 0→1 products with production-minded architecture from day one — shipped fast, without cutting the corners that matter.
004 Notes & deep dives
Task arithmetic on shared base checkpoints, decoder-weight preservation, blend-searching, and informal evaluation results from merging TTS models. A study in original experimentation over known patterns.
005 How I work
Most projects don't arrive as clean specs — they arrive half-formed, overloaded, or aimed at the wrong problem. The work starts by cross-questioning the brief to find what's actually needed, then simplifying the path to something that can be built and used in the real world.
Early-career but unusually high-output — production AI systems, workflow tooling, internal platforms, and writing-led products. Shipped and maintained, not prototyped and abandoned.
The first requirement is rarely the real one. I cross-question early to find what you actually want.
Push back, reframe, and cut friction before a line of code is written.
If something's unfamiliar, I take the time to learn it properly instead of bluffing.
Many of the best decisions show up during implementation, not before it.
Ship fast, stay close to the real workflow, and carry it until it holds up for real users.
Krishna Purwar
007 Let's talk
Open to AI product work, internal tools and workflow automation, full-stack AI systems, and selective consulting or roles.
No scheduler connected — this is a design placeholder.