Pipelines over
flashy demos.
I'm a Computer Science undergrad at JECRC, Jaipur, working across the full data stack — Data Engineering, Data Science, and Analytics. I like turning messy inputs into systems that run themselves.
My philosophy is simple: clean data over clever models, and bulletproof pipelines over flashy demos. Whether it's a medallion lakehouse or an autonomous content factory, I care about the thing still working tomorrow.
Technical stack
Python
FastAPI
Next.js
React
TypeScript
PostgreSQL
Redis
Docker
TensorFlow
Scikit-learn
OpenCV
Pandas
Apache Spark
Airflow
Jupyter
Git
GitHub
VS Code
Flask
Supabase
Google Cloud
SQLite
Linux
VercelFeatured work
End-to-end projects spanning data engineering, AI/ML, and automation — all real, all shipped.
TypeScriptPixiJSVitePWACog & Cosmos — Idle Game
A pixel-art interconnected incremental/idle game. Eight worlds feed one brass Fortune Engine — and unlike most idlers, every stage stays alive and feeds the others.
- Eight interconnected stages (Village → Multiverse) where labor, grain, ore, power, mana, chronons & shards compound across the chain
- 5 prestige/meta layers (Ascension, Transcendence, Reality Reset, Challenges), a global skill tree, achievements & relic collections, on break_eternity.js big-number math
- Installable offline-first PWA: PixiJS pixel scenes, IndexedDB autosave, optional Supabase cloud sync, synthesized Web-Audio SFX
Python
GeminiSupabaseFFmpegfaster-whisperedge-ttsYouTube APIGitHub ActionsAI Reel Factory — Autonomous Shorts
A near-$0/month autonomous content factory: it researches news, writes scripts, narrates, renders captioned vertical video, and publishes 4–5 YouTube Shorts/day — with one human Telegram tap per day.
- Idea→script→voice→captioned-video→publish pipeline orchestrated by GitHub Actions cron; one Telegram 'Morning Digest' approval per day
- Gemini (Groq failover) scripting, edge-tts narration, Pexels/Pixabay visuals, FFmpeg assembly, faster-whisper word-by-word captions
- Supabase Postgres state; publishes to YouTube @butitmatters at $0/month beyond an existing subscription
Next.jsFastAPIPythonNeon PostgresUpstash RedisGroq LLMSkill Issue — GitHub Intelligence
An AI-powered GitHub intelligence platform. Drop in a username and it turns repos, OSS contributions, and coding discipline into a deterministic engineering score and shareable receipts.
- Deterministic 100-point engineering score across repo quality, maturity, OSS, consistency & recruiter signal — AI only narrates
- FastAPI + Next.js with Neon Postgres persistence, GitHub OAuth, and an Upstash Redis warm cache (repeat analysis p95 ≤ 200 ms)
- Groq llama-3.3-70b Roast + Mentor narration and shareable 1200×630 'GitHub Receipts' OG scorecards
PythonApache AirflowPostgreSQLPower BIETLCRM + Sales — Data Warehouse
An end-to-end data warehousing project using the Maven Analytics dataset. Built a robust ETL pipeline that cleans and transforms raw CRM data into a structured PostgreSQL warehouse.
- Architected an automated ETL pipeline with Airflow to ingest and process 10k+ sales records
- Designed a star-schema warehouse in PostgreSQL for optimized analytical querying
- Surfaced actionable business insights via a dynamic Power BI dashboard focusing on sales performance
FastAPINext.jsNeon PostgresYouTube Wrapped — Data Pipeline
A personal 'Spotify Wrapped' for YouTube. End-to-end data pipeline that transforms Google Takeout exports into a polished year-in-review analytics dashboard.
- Medallion lakehouse pipeline (Bronze → Silver → Gold) in Databricks with Delta Lake tables
- FastAPI backend on Render serving analytics from Neon Postgres fact tables
- Next.js dashboard with animated cards, genre splits, binge sessions & listening rhythm charts
PythonLocal LLMsSemantic MemoryWake-word DetectionTTSJARVIS — Offline AI Assistant
A modular, privacy-first AI voice bot built with local LLM inference and semantic memory. Operates entirely offline without external APIs.
- Deployed a locally trained ML model for intent classification without external AI APIs
- Implemented speech recognition, intent classification & system automation in one integrated pipeline
- Built on Python + TensorFlow + NLP for real-time command processing
PythonFlaskOpenCVSQLiteSSEFaceFilter AI — Facial Recognition
A locally-run face-recognition platform that detects, matches, and organizes photos by face — no cloud uploads, no API keys.
- YuNet face detection + SFace 128-dimensional embeddings (ONNX) with cosine-similarity matching at a configurable threshold
- Caches pre-computed embeddings keyed on file hash to eliminate redundant inference
- Real-time progress streaming via Server-Sent Events; three-table SQLite schema for resumable runs
Next.jsTypeScriptSupabaseTailwind CSSTwilioMapLibreSahara — Women Safety App
A mobile-first safety web app with gesture-based SOS, real-time location tracking, and Twilio-powered alerts to trusted contacts — all backed by Supabase.
- Gesture-based SOS trigger with real-time location tracking via MapLibre GL
- Supabase (Postgres + auth) backend with row-level security and PLpgSQL functions
- Twilio-powered SMS alerts dispatched to saved trusted contacts on emergency
Internship journey
- Built ML classification and prediction models using Python with real-world datasets
- Cleaned, analyzed & engineered features to extract key business insights from raw data
- Worked on practical projects leveraging Scikit-learn, Pandas, and data visualization libraries
- Designed fully responsive web pages using HTML, CSS, and Bootstrap framework
- Improved UI/UX through clean, accessible layouts and user-centered design principles
- Delivered polished frontend interfaces for client-facing products
Education
and Research Center
Certs & achievements
Get in touch
Have a project, an opportunity, or just want to say hello? I'd love to hear from you.