I embed with real workflows — patent analysts, legal teams, plant engineers — and replace the slow manual parts with agents, RAG and document AI that hold up in production. No demos that die on real users: every system below has numbers behind it.
📂 Open the full case files
Junior Data Scientist — SciTech Patent Art Services · Hyderabad · Mar 2026 – Present
- Patent extraction pipeline (US/EP/JP): OCR + non-OCR with Azure OCR, PP-DocLayout / Docling (CPU/GPU-optimized) and Gemini bounding boxes for image cropping and LaTeX formulas, line-level coordinate mapping, real-time WebSocket collaboration → 3–5× lower cost. Re-designed server-to-server transfer to be fully async with auto-matching.
- Chart & claim construction: replaced a manual Adobe Acrobat workflow (30–40% accuracy, hours per document) with an automated pipeline — CPU-only OpenCV rectangle-merge for SMask images, multi-agent text-spillover detection, citation extraction → 95%+ accuracy, 90–95% lower cost, thousands of pages in minutes.
- Non-table claim charts: exact per-line coordinates across wildly varying layouts at $0.80 / 1,000 pages.
- Legal RAG chatbot: parallel ingestion + leaner agent orchestration → lower latency and token spend (FastAPI, PostgreSQL, LangChain/LangGraph).
- Website-to-PDF & legal media: Selenium full-page PDF engine that fixes broken Print-to-PDF; video/GIF pipeline with keyframe dedupe (24/30 FPS), Whisper transcripts and LLM summaries.
AI Software Engineer L1 — Gyan Data · Chennai · Aug 2025 – Mar 2026
- Telemetry AI dashboard for real-time turbine monitoring — autonomous agents, natural-language queries, streamed tokens, live Plotly (Streamlit).
- Graph-RAG backend with FastAPI + LightRAG + Neo4j and domain-constrained graph selection; Frappe ERP customization with Playwright E2E.
- DB-agnostic AI chatbot — LangChain/LangGraph, Google Auth, pluggable OpenRouter / Gemini / OpenAI / Ollama, LangSmith monitoring.
Data Science Intern — Gyan Data · Chennai · Jan 2025 – Jul 2025
- ERP customization + AI chatbot development → direct full-time conversion offer.
Atrophy on PyPI · Atrophy site · Deep Research walkthrough · NeurIQ AI live · KV Nexus live
🗄️ From the earlier shelves
| Project | Stack | What it does |
|---|---|---|
| AI Flowchart Generator | Flask · GenAI · Python | Natural language → flowcharts, instantly |
| Intelligent Doc Summarizer | LLM optimization · NLP | Condenses large documents with precision |
| AYUSH Herbal Garden | Flask · Firebase · A-Frame 3D | Immersive VR/3D garden of 50+ medicinal plants with GSAP animations |
| AI Text Classifier | Python · NLP | Detects AI-generated vs human-written text |
| Zomato Restaurant App | Full stack · Analytics | Restaurant data analysis & search platform |
off the clock: ♟️ chess · 📷 photography · 🧮 competitive programming · speaks English, Telugu, Hindi
every pixel above is hand-written SVG · zero claims without artifacts · © 2026 Karri Vamsi Krishna


