Applied ML · computer vision · automation
Models that grade 90k sheets — not another chatbot wrapper.
Deterministic CV over a learned bubble detector. Ambiguous marks go to review. Inventory image recognition. Scraping that drives a decision, not a demo notebook.
- Warp and persist — a bad scan does not redo the batch.
- Align to printed marks before any density call.
- Relative histogram, not a GPU detector.
- Ambiguous marks go to a human queue. Silent misgrade is worse.
Why this desk, not a generic portfolio
| Typical | Here |
|---|---|
| Fine-tune a blog generator | OMR pipeline: warp → align → density → rules → queue |
| Unverified latency claims | Reproducible scores, no GPU, institutional exam constraint |
| Prompt playgrounds | RAG and Hopfield work documented as experiments, labeled as such |
Proof for this desk
OMR Sheet Automation
Built a staged, fault-tolerant Python OMR pipeline that processed over 90,000 physical examination forms.
RAG Document Chatbot
Retrieval-augmented chatbot over PDF docs: embed with a hosted model, retrieve top chunks, answer with an LLM constrained to that context — Flask UI, honest notes on global-index limits.
HTML Content Translator
Python utility that walks an HTML DOM with BeautifulSoup, translates visible text nodes (EN→HI), and leaves tags, attributes, and script/style alone.
Hopfield Network Visualization
Two Jupyter notebooks exploring Hopfield networks — a basic 2-neuron energy landscape demo and an advanced 5x5 grid showing pattern storage and noisy-input recovery with animation. Built with AI assistance while learning the concepts.
Campus Exam Portal
Built Campus Quest, a campus examination portal with per-session question/option permutation, focus-loss integrity signals, and bulk question-bank ingestion for faculty.
E-commerce Price & Catalog Sync
Automated supplier catalog extraction and price-drift detection for an electronics e-commerce store — margin rules and outlier thresholds produce OpenCart import manifests instead of manual daily price checks.