naga adarsh
cv ↓
open to sde · full-stack · ai/ml roleschennai, in · class of 2027

naga adarsh

Final-year IT undergrad in Chennai, graduating 2027. I build full-stack and AI systems in Python and React — a MobileNetV2 freshness model behind a published patent application, a FAISS and ColBERT retrieval engine, and a Redis-cached URL shortener.

see the projects ↓email me →

every claim below is a link you can open and check.

selected work

Three things I built end to end, with the numbers they actually hit.

01computer vision · patent

AI-Based Freshness Indicator

Point a camera at fruit and vegetables, and it scores how fresh they are — the model behind patent application IN202641004962 A1.

  • Trained a MobileNetV2 CNN on 3,000+ fruit and vegetable images, scoring 92% F2 on the evaluation set
  • Added shelf-life decay modelling — the score falls with age and triggers a spoilage alert
  • Deployed behind a Flask API with OpenCV preprocessing and a React upload page
92% F2 · 3,000+ images
view code →live demo →patent IN202641004962 A1 →
pythontensorflowkerasopencvflaskreact
02retrieval-augmented qa

Scholar RAG Engine

Ask a research question in plain language, get an answer drawn from the papers you chose.

  • Combined FAISS candidate retrieval with ColBERT-style late-interaction scoring over research papers
  • Added a cross-encoder reranker, which raised answer precision by 30–40% in my evaluation
  • Served it with FastAPI; the Gemini API writes answers from the retrieved passages
+30–40% answer precision
view code →live demo →live demo may cold-boot
pythonfastapifaisscolbertsentence transformers
03full stack + caching

Nag-Link URL Shortener

A URL shortener with custom slugs — a React and TypeScript front end over a FastAPI service, cached in Redis.

  • Built the React and TypeScript front end on a FastAPI service with Upstash Redis
  • Redirects resolve in under 100 ms on Vercel; Redis caching cut lookup time by about 60%
<100 ms redirects
view code →live demo →
reacttypescriptfastapiredisvercel

train a tiny neural net

A real mini network, like the one inside the freshness project, learning to tell two classes apart. Each preset is one experiment — pick it and press run.

8 neurons with light dropout — the well-behaved setup.
  1. two kinds of dots — two classes. imagine fresh fruit vs spoiled fruit.
  2. the shaded zone — where the network thinks class 1 lives.
  3. training score — how well it remembers the examples it trained on.
  4. brand-new score — how it does on fruit it has never seen. that one matters.
  5. the lit circles — its decision cells. drop one and it has to learn with less.
xy0/1

8 lit neurons in the middle — click one to drop it out

class 1class 0got it wrongnetwork says class 1
training done0 steps
training data score—
brand-new data score—
not trained yetpress “run the full experiment” and watch the boundary appear.

experience & college

Two web roles, paid delivery, and the degree that runs alongside them.

sep 2025 — oct 2025 · freelance · chennai

Freelance Web Developer — SubbaiH Tuition & Daycare

Built a five-page site in HTML, CSS and JavaScript for a Chennai tuition and daycare centre (home, about, services, blog, contact). Handled requirements and delivery directly with the owner as a ₹10,000 paid engagement, then deployed it to static hosting and handed over written documentation.

see the project documentation →
html · css · javascriptfive pagesstatic hostingowner handover
jun 2025 — jul 2025 · chennai

Web Developer — Eagle Hi-Tech Softclou Pvt Ltd

  • Gathered client requirements and built three sites — two responsive HTML/CSS sites and one React app.
  • Built React components with state management and CSS layouts, then checked cross-browser output.
react.jshtml · cssjavascriptcross-browser testing
2023 — 2027 (expected)

B.Tech in Information Technology — Easwari Engineering College

Chennai. Most of my projects and coursework sit in the full-stack, AI/ML and backend space, with data structures and algorithms as the base.

GPA 8.3/10chennai, in2023 — 2027 (expected)

credentials

Certs, contest finals and coding scores — the links go to the real thing.

patentPublished patent application IN202641004962 A1 — AI-based freshness detection
hackathonSmart India Hackathon — college-level finalist, 2024 & 2025
certificationMicrosoft Certified: Azure Developer Associate
certificationCisco CCNA 1 · 2 · 3 — Networking Academy

competitive programming

730+problems solved on LeetCode1700+CodeChef rating, division 2870+problems across platforms on Codolio92%F2 on 3,000+ freshness images

more on github