SKIP TO CONTENT

SOFTWARE DEVELOPER · AI/ML ENGINEER

Building intelligent systems end-to-end.

B.TECH CSE '25LUCKNOW, INDIAOPEN TO SWE / AI-ML ROLES

[02]ABOUT

BOTH SIDES
OF THE STACK

I ship the product — and train the model behind it.

Web products with Next.js and TypeScript on one side; the models that make them intelligent — PyTorch, computer vision, retrieval-augmented LLMs — on the other. My flagship, Prism, is a deployed news-transparency platform that grounds LLM fact-check verdicts in live web citations.

Underneath it all: a solid DSA foundation in C++ and Python, 300+ LeetCode problems deep, and a habit of building systems that degrade gracefully instead of crashing.

EDUCATION

Lovely Professional University

Phagwara, Punjab, India

B.Tech in Computer Science Engineering

GRADUATED — JULY 2025

Data Structures & AlgorithmsOperating SystemsDatabase Management SystemsObject-Oriented Programming

COMPETITIVE

LEETCODE

300+ PROBLEMS SOLVED

1600+ CONTEST RATING

300+

LEETCODE PROBLEMS SOLVED

1600+

LEETCODE CONTEST RATING

15+

PROJECTS ENGINEERED

2

PLATFORMS LIVE IN PRODUCTION

C++PYTHONJAVASCRIPTTYPESCRIPTSQLREACTNEXT.JSTAILWIND CSSHTMLCSSNODE.JSEXPRESSREST APISJWT AUTHPYTORCHOPENCVYOLOV8DEEP LEARNINGCOMPUTER VISIONLLMSRAGPROMPT ENGINEERINGMONGODBMYSQLREDISGITGITHUBDOCKERCI/CDVERCELAWSPOSTMANLINUX

[03]STACK

INSTRUMENT PANEL

[03.1] LANGUAGES05

  • C++
  • Python
  • JavaScript
  • TypeScript
  • SQL

[03.2] FRONTEND05

  • React
  • Next.js
  • Tailwind CSS
  • HTML
  • CSS

[03.3] BACKEND04

  • Node.js
  • Express
  • REST APIs
  • JWT Auth

[03.4] AI / ML08

  • PyTorch
  • OpenCV
  • YOLOv8
  • Deep Learning
  • Computer Vision
  • LLMs
  • RAG
  • Prompt Engineering

[03.5] DATABASES03

  • MongoDB
  • MySQL
  • Redis

[03.6] TOOLS08

  • Git
  • GitHub
  • Docker
  • CI/CD
  • Vercel
  • AWS
  • Postman
  • Linux

[04]WORK

PROOF OF WORK

THREE FLAGSHIP SYSTEMS — ONE IN PRODUCTION, ONE WATCHING VIDEO STREAMS, ONE READING THE SUN. THE ARCHIVE HOLDS THE REST.

001

PRISM

FLAGSHIP · LIVE

NEWS TRANSPARENCY & BIAS ANALYSIS PLATFORM

Refracts any article, YouTube link, or pasted text into a six-stage transparency analysis — claim classification, source provenance, live fact-checking, and cross-outlet coverage gaps.

  • Deployed full-stack platform: sentence-level claim/opinion/rhetoric classification, source forensics, and coverage-gap comparison across outlets.
  • RAG fact-checking pipeline ranks an article's most salient claims, searches them against live web sources, and grounds Llama 3.3 70B verdicts in real citations — with a ≥2-source backfill so no verdict rests on a single proof.
  • Resilient by design: fail-open Redis caching, per-IP rate limiting, keyless RDAP/DNS/IP forensics and YouTube transcript extraction — degrading gracefully when a dependency is absent.

TELEMETRY

Next.jsTypeScriptReactTailwindNode.jsGroq · Llama 3.3 70BRAGRedis
002

SENTINEL VISION

REAL-TIME SURVEILLANCE THREAT DETECTION

A multi-stage computer-vision pipeline that watches video streams for conflict, weapons, and crowd risk — scoring threats in real time.

  • Chains a Swin Transformer conflict classifier, a fine-tuned YOLOv8 weapon detector, and a crowd-density estimator into one unified threat-scoring engine.
  • Trained on a custom-labelled surveillance dataset with targeted augmentation and class balancing to cut false positives on crowded, visually ambiguous scenes.
  • GPU-accelerated CUDA inference loop decodes, preprocesses, and scores frames in real time, overlaying live alerts on streams and recorded footage.

TELEMETRY

RECALL83%
PRECISION75%
F1 SCORE0.79
PythonPyTorchYOLOv8Swin TransformerOpenCVCUDA
003

SCENIC VIEW ORACLE

RAIL JOURNEY LIGHT & SEAT-SIDE PLANNER

Answers one very specific travel question — which side of the train to sit on, and when to look out the window — by computing it from route geometry, published schedules, and real-time solar position instead of static rules.

  • Derives a left/right seating verdict per journey from route geometry, timetable data, and astronomical sun position, attaching a confidence level and readable reasoning to every recommendation.
  • Renders a Light Band across the whole journey — sun trajectory, lighting conditions, and the exact time windows and viewing directions for each scenic stretch, mapped down to berth groups and window walls.
  • Ten hand-verified Indian rail routes including the Darjeeling Himalayan, Kalka–Shimla, and Nilgiri Mountain railways, built on OpenStreetMap and open datasets with no scraping and no API keys, and covered by property-based and end-to-end tests.

TELEMETRY

Next.js 15React 19TypeScriptTailwind v4MapLibre GLAstronomy EngineZodPlaywright

ARCHIVE

SMALLER BUILDS & EXPERIMENTS — SOURCE ON GITHUB.

2026

WEAPON DETECTION

Real-time YOLO weapon detector for security feeds — a modular camera / inference / alert pipeline that runs on CPU and accelerates on CUDA.

PythonPyTorchYOLOOpenCV

[05] CONTACT — TRANSMISSION OPEN

LET'S BUILD
SOMETHING.

Looking for a developer who can ship the interface and train the model? My inbox is open.