Anup Bhat is a computer engineer and AI developer from Goa, India. He builds machine learning systems, healthcare technology, language-model products, civic-tech tools, and interactive web products. He has won multiple hackathons including HackIndia Goa Regional (1st Place) and was a Smart India Hackathon 2024 finalist.

What projects has Anup Bhat built?

Anup Bhat has built Dark Store Map (mapping quick-commerce infrastructure across Indian cities with 8K+ users), Xpose (a browser-based GitHub repository explorer), Predicting Non-Communicable Diseases from Retinal Fundus Images (multi-label retinal disease screening), Eikasia (a cinematic browser image editor), Chesume (an open-source AI interview preparation platform), NetOps Agent (an AI-assisted network change workflow tool), ArthGyan Financial Literacy, MnemoPlay (a gamified learning app with AR), Financial FAQ Chatbot using MeTTa knowledge graphs, EWET Nutrition App, and several other AI and civic-tech projects.

What are Anup Bhat's technical skills?

Anup Bhat's technical skills include AI/ML (PyTorch, Transformers, PEFT, Diffusers, Hugging Face, LangChain, LangGraph, Pydantic-AI), Python (NLTK, Spacy, Flask, FastAPI, Django, OpenCV, Pandas), Web Development (React.js, React Native, Next.js, Express, Node.js, Flutter), Cloud & Data (AWS, Docker, Google Cloud, Vercel, Databricks, MongoDB, PostgreSQL, Firebase, Redis, Supabase), and Languages (JavaScript, TypeScript, Python, C/C++, SQL, R, Dart).

What hackathons has Anup Bhat won?

Anup Bhat has won 1st Place at HackIndia Goa Regional, 1st Place at Infofest Goa University, 1st Place at DBCE Inspirathon, 1st Place at Pixel Play PCCE. He was also a Smart India Hackathon 2024 Finalist, placed 2nd at Techfluence GCE and FlutterFlow Hackathon GDG Goa, 3rd at Puch AI Hackathon (out of 800 teams), and was a finalist at Goa Police Hackathon 2025, Build India (Anthropic x Replit x Lightspeed), and MESA x Emergent Labs Hackathon.

What is Anup Bhat's portfolio website?

Anup Bhat's portfolio website is anupbhat.com. It showcases his projects, blog posts on AI and engineering, achievements, technical skills, and a curated 'Notes to Self' section with resources on neural networks, philosophy, and builder mindset.

Anup Bhat - AI Engineer, Developer, and Builder

Building at the intersection of nature, cognition, and intelligent systems.

I'm Anup Bhat, a computer engineer from Goa, India. I have a passion for building at the intersection of nature, cognition, and intelligent systems. My interests span biomimicry, machine learning, artificial intelligence, and the psychological underpinnings of human–machine interaction. I'm particularly drawn to healthcare technologies and the development of language models, where I see immense potential for impactful, inclusive innovation.

Whether it's designing AI that learns from biological systems or creating tools that serve real-world needs, I'm driven by a constant desire to solve complex problems through thoughtful, interdisciplinary engineering. I'm also into Open Source Intelligence.

“What I cannot create, I do not understand.” - Richard Feynman

Selected work

Each project began as discomfort with how the world currently works.

Dark Store Map

Reached 8K+ active users by mapping quick-commerce dark store coverage for Blinkit, Zepto, and Instamart across five major Indian metros.

Leaflet, GeoJSON, OpenStreetMap

Reached 8K+ active users by mapping quick-commerce dark store coverage for Blinkit, Zepto, and Instamart across five major Indian metros.
Improved delivery-footprint realism by generating 10-minute coverage zones with Geoapify isolines, road networks, and city-specific rider speed assumptions.
Made city-level infrastructure easier to compare by building Leaflet heatmaps, polygon views, metro layers, and neighborhood-level delivery context.
Reduced map load friction by precomputing high-fidelity polygon data for instant interactive browsing.

Live Map

Xpose

Made large codebases easier to inspect in the browser by building a repository explorer for public GitHub repos and ZIP uploads with syntax highlighting across 10+ languages.

FastAPI, Next.js, TypeScript, Prism.js, tiktoken, httpx

Made large codebases easier to inspect in the browser by building a repository explorer for public GitHub repos and ZIP uploads with syntax highlighting across 10+ languages.
Reduced initial DOM nodes by 95% using Intersection Observer and requestAnimationFrame virtualization for smooth large-repo navigation.
Improved AI-context preparation by adding tiktoken budgeting for selected files and repository slices.
Hardened repo ingestion with host allowlists, binary-file heuristics, and ZIP path-traversal validation.

Live App

Predicting Non-Communicable Diseases from Retinal Fundus Images

Achieved 0.9441 AUC for multi-label retinal disease screening by ensembling EfficientNet-B3 and MaxViT-Tiny on fundus image datasets.

PyTorch, EfficientNet-B3, MaxViT-Tiny, OpenCV, CLAHE

Achieved 0.9441 AUC for multi-label retinal disease screening by ensembling EfficientNet-B3 and MaxViT-Tiny on fundus image datasets.
Improved feature consistency by unifying DR and RFMiD data with supervision masks, LAB-space CLAHE preprocessing, ImageNet normalization, and online augmentation.
Strengthened rare-disease learning by combining class-balanced BCE, label smoothing, and RFMiD-weighted sampling.
Validated model choice by benchmarking Inception-V3, EfficientNet-B3, ConvNeXt-Tiny, and MaxViT-Tiny, with EfficientNet-B3 reaching 0.9179 macro AUC.

Eikasia

Enabled creators to produce cinematic PNG/JPEG exports in-browser by building a client-side image editor with film looks, color controls, typography, overlays, and crop tools.

Next.js 16, React 19, TypeScript, Tailwind CSS, Radix UI, Fabric.js, OffscreenCanvas

Enabled creators to produce cinematic PNG/JPEG exports in-browser by building a client-side image editor with film looks, color controls, typography, overlays, and crop tools.
Preserved export fidelity by rendering from typed project state through OffscreenCanvas instead of screenshotting the DOM.
Recreated film-style looks with color matrices, blend modes, grain, vignette, halation, fade, sharpness, and adjustment controls.
Made complex edits reversible by designing a typed editor model with look presets, Fabric.js text layers, responsive coordinate mapping, and 50-step undo/redo history.
Improved reliability for large uploads by adding PNG/JPEG/WEBP handling, object URL cleanup, size validation, responsive canvas fitting, and max-dimension export capping.

Live Editor

Chesume

Personalized technical interview prep by building an open-source platform that parses resumes and generates tailored question kits, answers, and study material.

Next.js, React, Gemini, TypeScript, Tailwind CSS, Vercel

Personalized technical interview prep by building an open-source platform that parses resumes and generates tailored question kits, answers, and study material.
Made resume claims harder to bluff by extracting project and tech-stack signals, then generating probing follow-up questions for each candidate.
Reduced random interview prep by organizing the Golden 20 algorithm patterns, 80+ practice questions, interview puzzles, guided practice flows, and blog resources.
Improved learning flow across devices with a mobile-friendly interface for resume scans, tailored kits, pattern-based DSA resources, and quick starts.

Live Platform

NetOps Agent

Reduced risk in network configuration work by building an assistant that converts operator requests into inspectable plans, waits for approval, executes automation, and returns validation evidence.

FastAPI, Next.js, LangGraph, Netmiko, OpenAI, Pydantic

Reduced risk in network configuration work by building an assistant that converts operator requests into inspectable plans, waits for approval, executes automation, and returns validation evidence.
Kept AI-assisted changes controlled by using AI for intent understanding while the backend validates actions and executes only approved operations.
Demonstrated realistic workflows by connecting the demo to a sandboxed Cat8000 router for read-only checks, routing table inspection, static route changes, and OSPF workflows.
Separated planning from execution with a FastAPI backend for sessions, planning, execution, and validation plus a Next.js workflow console.
Improved operational auditability by preserving structured plans, approval checkpoints, automation logs, validation output, and proof of what changed.

GitHub

Tools I reach for when the terrain changes.

So sometimes I get lucky.

HackIndia Goa Regional

1st Place

Led the 1st place team out of 70 regional teams and represented Goa at nationals in Delhi.

Infofest Goa University

1st Place

Led the team to 1st place in the Goa University tech fest.

DBCE Inspirathon

1st Place

Led the team to 1st place in the DBCE hackathon.

A map for you to see I touch grass.

GitHub Twitter LinkedIn Email

Writing at the edge of code and consciousness.

01The secret to 10 minute delivery isn't driver speedThe secret to 10 minute delivery isn't driver speed. The average delivery distance is shorter than your morning walk. Dark stores are everywhere, hiding in pl

02The Sovereign StackReading is not data entry. It is a firmware update. I used to believe that the primary metric for reading was throughput (pages per hour). My implicit assumptNovember 20, 2025

03Your Brain on ChatGPTWhat happens when we outsource our thinking to AI? A recent MIT study holds a sobering mirror to our cognitive habits. The Brain Scans Don't Lie Researchers sAugust 12, 2025

Hi! I am Anup Bhat

Building at the intersection of nature, cognition, and intelligent systems.