Hariprasad BDesign Engineer
I'm a product designer who trained as an engineer. I design the work and ship it, Figma to React, without losing the thread between them, and I'm always building something of my own on the side.
I think before I open a tool. Rough ideas first, in my head or on paper, then research, then code. The gap between a wireframe and a shipped product is where most things fall apart. I try to close it by staying in both lanes.
Spent time doing UI/UX at Metastart and now designing enterprise B2B iPaaS at UnifyApps via ICD.
The ideal scenario is you don't outsource thinking, you outsource execution. That's the thing AI makes easier to get wrong.
Open retrieval standard. Hybrid sparse retrieval with .VCB corpus bundles. Cuts inference token load by 97%. Built on 8 papers, MIT license.
Token-optimized site maps for AI agents. A compact JSON map that cuts 85–95% of the tokens a sitemap costs. This very site ships its own.
Wedding sites in two minutes. No code, free, wildcard subdomains. Built on Next.js, Firebase, and Vercel.
Compute belongs where it compounds.
Most AI systems treat every query as if the answer doesn't exist yet. It usually does. The problem is retrieval, not generation. Throwing more inference at bad data delivery is a physics problem dressed up as a software one.
I didn't want to just hold that opinion, so I attempted to build Context Engine, an open retrieval standard with 97% fewer tokens per query, 8 papers, MIT licensed. The goal was always to make the expensive part smaller, not faster.
See all ResearchesMountains and forests when I can. Camera and sketchbook when I can't. Music when neither is possible.