Skip to content
← back to portfoliorésumé ↗
About

Profile, uncompressed.

I build systems that make AI feel physical — voice that lands on a real device, retrieval you can interrogate, interfaces that show their working. Then I design and ship the surface around them, because a model without a UI is a demo.

This page is the part a résumé cannot hold: how I decide, what I refuse to round up, and what is still open. The chronology lives at /resume; the reasoning lives in the case files.

Operating principles

five rules
  1. p01

    Ship it on the device

    A build that only survives on my laptop is a demo. Voice, retrieval and vision work has to hold up on a real phone, a real microphone and a real network — so that is where I test it first, not last.

  2. p02

    Measure before claiming

    Latency numbers over adjectives. If I say something feels instant, there is a measurement behind the sentence — and if there isn't one yet, I say so instead.

  3. p03

    Show the working

    Readable state beats a clever black box. That is why this site has a case terminal, why every dossier lists what went wrong, and why the unfinished parts are on the page.

  4. p04

    The screen is part of the system

    A model without an interface is a demo, and an interface without an understanding of the model is theatre. I try to own both halves of the same problem.

  5. p05

    Reuse what is already proven

    Before inventing a mechanism, I look for the one that already works — in the standard library, in the platform, or in a script I wrote months ago. Novelty is not a feature.

  6. status

    Open to selected builds where the interface and the model are the same conversation.

    route a signal →

Working surface

Systems
Python · TypeScript · C · Java
Web
Next.js App Router · React · Tailwind
Speech & vision
Whisper · on-device STT · TTS pipelines
Retrieval
Embeddings · RAG · semantic note graphs
Native
macOS · Android · desktop automation
Design
Figma · design systems · motion specs

Path so far

  • NowVoidLabs — UI/UX design & frontend lead2026 — present
  • NowCrossStack — full-stack, community platform2025 — present
  • StudyingB.Tech Computer Science, Gitam2025 — 2029
  • Long-runningTeja Visual Arts — edits, films, visual work2020 — present

What is not finished

  • Recognition for code-switched Telugu-and-English speech in Chunks is still behind the quality of what it produces on screen.
  • iTantra has documented gaps between what the design promised and what the build delivers; I have not closed them.
  • Most of these repositories are private source, so the case files carry reasoning instead of proof links.

The hero of this site promises an honest list of what is not done yet. That promise is only worth keeping if the list is real, so it stays here and gets shorter over time.

This site is a build

  • Next.js 15 · App Routerframework
  • three.js particle nameplatehero
  • framer-motion, gated on reduced-motionmotion
  • 175 kB first-load JSweight
  • AA contrast floor, keyboard-complete case browseraccess
  • Print-ready résumé at /resumeoutput
Elsewhere
rishideep.mandavilli@gmail.com→
GitHub ↗LinkedIn ↗YouTube ↗X ↗Instagram ↗
Case files →Résumé ↗