builds

engineered solutions & technical projects.

floaty

desktop-ai

context-aware desktop assistant operating as a keyboard-summoned floating pill to automate developer workflows. obtained 50+ downloads on launch day.

  • engineered cross-platform desktop agent in electron with low-overhead memory footprint (<80mb).
  • developed isolated local execution loop for python/nodejs scripting delivering context-aware terminal completions with 40% lower latency.
  • designed custom multi-provider api router with rate-limit fallbacks & vision token compression, reducing cloud llm costs by 35%.
stack: electron, python, nodejs, multi-provider-router

rag-based conversational vision model for document analysis over 100+ page pdf documents with 94% accuracy.

  • programmed advanced rag engine utilizing vector embeddings and hybrid search.
  • engineered low-latency voice interactivity (stt/tts) achieving conversational audio feedback loops under 250ms.
  • built cross-platform web client alongside native gtk desktop ui with high-performance file-io streams.
stack: python, rag, vector-embeddings, gtk, stt/tts

zero-latency client-side timetable parser and interactive web application for jaypee university of information technology.

  • engineered in-browser excel parsing engine using sheetjs (xlsx) to extract multi-sheet timetables entirely client-side.
  • developed regex pre-cleaning & batch range expansion algorithms to parse corrupted spreadsheet formats and complex batch configurations.
  • built serverless backend routes with supabase postgresql for crowdsourced subject data and tesseract.js ocr screenshot parsing.
  • implemented sliding-window ip rate limiting and client-side throttling to protect report & contribute endpoints.
stack: nextjs, typescript, sheetjs, supabase, tailwindcss, radix-ui, tesseract.js