I build asynchronous backend services, document intelligence pipelines, and local-first RAG architectures. Graduated in Computer Science (Batch of 2025) with a strong foundation in core algorithms and systems. The part of engineering I enjoy most is exploring new ideas, diving deep into how tools work under the hood, and building projects that push me to learn and grow both technically and personally.
| Focus | Document intelligence (two-phase OCR & orientation correction), low-latency hybrid RAG, agent toolchains, and async backend services. |
| Currently focusing | Deepening systems design, sharpening core algorithms and backend stack depth, and mastering AI pair-programming workflows. |
| Education | B.E. Computer Science and Engineering — Batch of 2025. |
backnotprop/plannotator
Annotate and review coding agent plans and code diffs visually, share with your team, send feedback to agents with one click.
- #1669 fix(ai): show friendly error when opencode CLI is unavailable
- #1658 fix(ui): style Ask AI dropdown options for dark themes
QwenLM/qwen-code
An open-source AI coding agent that lives in your terminal.
- #1890 fix(windows): resolve silent failures caused by CRLF line endings (#1868)
rtk-ai/rtk
CLI proxy that reduces LLM token consumption by 60-90% on common dev commands. Single Rust binary, zero dependencies
- #3711 feat(hook): add native Google Antigravity plugin lifecycle and hook support
3 open pull requests across 2 repositories
rtk-ai/rtk
CLI proxy that reduces LLM token consumption by 60-90% on common dev commands. Single Rust binary, zero dependencies
- #4343 fix(hooks): recognize configured agent hooks in warning check (#913)
- #4335 fix(init): migrate legacy Antigravity rules on install and uninstall (#4317)
backnotprop/plannotator
Annotate and review coding agent plans and code diffs visually, share with your team, send feedback to agents with one click.
- #1646 fix(plan): smooth sticky lane badge expansion and hover sync
Updated nightly via GitHub Actions with pr-showcase · all pull requests →
|
FastAPI pipeline for Indian identity document classification and OCR field extraction. Implemented two-phase OCR with YOLO-based orientation correction to eliminate text extraction noise on skewed scans. Achieved 98%+ classification accuracy across 7 document formats (50+ field classes) and ~80% latency reduction over naive whole-image OCR passes. FastAPI PyTorch YOLO PaddleOCR OpenCV Pydantic pytest
|
Privacy-first local RAG engine for multi-document querying. Built hybrid semantic + BM25 keyword search unified via Reciprocal Rank Fusion (RRF) to eliminate vector-only precision loss on precise domain identifiers. Features local ChromaDB chunk storage and switchable local/cloud inference pipeline with deduplication. FastAPI ChromaDB PyTorch HuggingFace Gemini API
|
MITRA Theatre client projectLive cinematic website built independently for a live theatre production organization. Engineered an automated local validation pipeline (Husky pre-commit/pre-push hooks, Knip dead-code detection, Conventional Commits) paired with Vitest and Playwright browser integration testing. React 19 TypeScript Vite Framer Motion Vitest Playwright
|
MindMate-AI team projectCollaborative AI mental health platform. Engineered the real-time voice pipeline (Twilio WebRTC + ElevenLabs TTS), asynchronous WebSocket consumers for low-latency session messaging, automated PHQ-9 clinical assessment workflows, and JWT authentication. Django Redis Channels WebSockets Twilio ElevenLabs
|
Open to full-time Python Backend Engineer, AI Engineer, and Software Engineer roles (Remote or On-site).
- Email: aayushyaash@outlook.com
- LinkedIn: linkedin.com/in/aayush-yash
- Portfolio: aayushyaash.github.io/portfolio
Fun fact: I enjoy music 
, anime 
, and reading light novels 