|
I bridge the gap between cutting-edge ML research and polished, production-grade full-stack systems — turning complex model architectures into real-world intelligent applications. My work spans seismic hazard intelligence, urban computer vision, industrial inspection automation, and climate-resilience AI. Name : Vansh Agrawal
Role : Full-Stack Developer & ML Researcher
Domain : Computer Vision · Urban AI · Structural Informatics
Location : Chennai, India (UTC +05:30)
Status : Building @ the intersection of research & deploymentCurrently researching a flood prediction model for the Bihar basin — applying hydrological ML to infrastructure resilience at regional scale. |
|
A global seismic hazard intelligence platform delivering real-time tectonic monitoring, probabilistic hazard scoring, and volumetric 3D tectonic simulations. Combines gradient-boosted ensemble models with interactive WebGL globe rendering. Architecture Highlights
|
An edge-ready visual intelligence framework that quantifies Urban Cleanliness Index (UCI) from municipal imagery. Combines real-time object detection with a generative augmentation pipeline for robust model training across diverse urban environments. Architecture Highlights
|
An enterprise-grade defect detection system for structural TMT steel bar manufacturing. Provides automated surface anomaly classification with sub-millimetre precision, reducing manual inspection overhead in high-throughput industrial pipelines. Architecture Highlights
|
A production FastAPI microservice exposing trained ML models for wine quality predictive analytics. Built with clean API design, schema validation, and async inference — serving as a canonical reference for deploying ML in enterprise REST architectures. Architecture Highlights
|
|
Multi-agent AI placement automation for B.Tech CSE students — autonomous job discovery, eligibility analysis, resume tailoring, and application tracking via a live dashboard. |
Real-time peer-to-peer WebRTC application for low-latency voice, video, and screen sharing — built for production-grade communication reliability. |
High-performance Retrieval-Augmented Generation platform. Upload PDFs and interrogate them via semantic search and LLM reasoning — minimal latency, maximum accuracy. |
|
Automated micro-insurance platform providing income security to gig-economy delivery workers via data-driven risk profiling and real-time payout automation. |
Real-time hand gesture recognition using MediaPipe landmarks + deep learning. Enables touchless HCI through accurate multi-gesture classification at high frame rates. |
Compact, explainable career recommendation engine mapping student profiles (skills, education, interests) to ranked career suggestions with transparent reasoning. |
🌊 ACTIVE RESEARCH · Flood Prediction Model — Bihar Basin
┌─────────────────────────────────────────────────────────────────────┐
│ │
│ Developing a hydrological ML framework for flood risk │
│ forecasting across the Bihar river basin network. The system │
│ integrates multi-source geospatial data (satellite imagery, │
│ DEM, river gauge telemetry) with temporal deep learning │
│ architectures to generate 72-hour probabilistic inundation │
│ maps — directly informing infrastructure resilience planning │
│ and early warning systems at the district level. │
│ │
│ Stack : Python · PyTorch · LSTM / Transformer · GIS │
│ Data : MODIS · Sentinel-1 SAR · IMD rainfall grids │
│ Goal : Sub-district flood extent prediction with ≥ 85% IoU │
│ │
└─────────────────────────────────────────────────────────────────────┘
Core Languages
ML & Computer Vision
Full-Stack & APIs
Data, Infrastructure & Tools
|
🎵 Music is the other language I speak. Beyond systems and models, I maintain an active social presence dedicated to singing — a creative counterbalance to the precision of engineering. Music, like good architecture, is about finding the right structure for something that needs to breathe. |
📍 Chennai, India | UTC +05:30 Building systems that matter — from tectonic faults to urban streets to flood plains. Every line of code is an attempt to make the physical world a little more legible to machines, and a little safer for people. Open to research collaborations, internships, and interesting problems. |
"The most elegant systems are those that hide their complexity behind a clean surface."
