Software engineer · Bengaluru

Hello there, I'm Ujesh.

Software engineer building production web apps, applied AI and distributed GPU systems, end to end. Seven hackathon wins, and a paper at IMPACT-2027.

Scroll to explore

WHERE THE WORK HAS LANDED

Two startups, an ML internship at Hitachi, fifty-plus hackathon floors and a conference stage in January 2027.

  • EaseMedInternship
  • EdactlyFreelance
  • HitachiML internship
  • AICTE × Microsoft & SAPML internship
  • InceptionWinner
  • DenovaWinner
  • CardanoWinner
  • Cypher 1 & 3Winner ×2
  • IMPACT-2027Paper accepted
  • MNNIT AllahabadBook chapter
  • LitmusChaosOpen source
  • Smart India HackathonCollege rank 1
  • HackerRankBronze
  • Google Solution ChallengeRegional qualifier
  • ICPCPrelims ×2

The story so far

Eight chapters, one habit: build the thing, then show up for it.

  1. #01College, the long way roundComputer science at Atria (VTU) since 2023, plus a 200-plus member code club I now lead.
  2. #02Fifty-plus hackathonsSeven wins, twenty national finals, and five more hackathons I organised for 500+ people each.
  3. #03InternshipsOutbreak prediction at EaseMed, bird-species detection at Hitachi, lightweight vision models with AICTE.
  4. #04FreelanceEdactly's prompt-to-lesson video pipeline, plus sites for NEXR, a resort and a merch store.
  5. #05Research & conferencesGradMesh at IMPACT-2027, a book chapter for Viksit Bharat 2047, and the next paper in progress.
  6. #06AI/ML, in the roomAI/ML events and talks, on the stage and in the audience. RAG, vision and GPUs, mostly.
  7. #07A business of my ownI've raised money to start a business. More on that soon.
  8. #08Off the clockMarathons, the gym every day, football, dancing and a lot of games.
Ujesh pitching with a microphone

01

Full-stack products

React, Next.js and Node, schema to deploy

Real-time messaging, authentication and role-based access, shipped to hospital and vendor teams at EaseMed.

The team on stage with certificates at Cypher

02

Applied AI & RAG

Models that earn their place in the product

RAG over the NCERT curriculum, embeddings, semantic search and LLM inference services running on AWS.

Ujesh with the Inception winner's cheque

03

Distributed GPU systems

Idle GPUs, one training cluster

GradMesh pools consumer GPUs across a LAN: up to 1.8× faster than a single node across four GPUs.

A packed hackathon floor

04

Real-time 3D & WebGL

Worlds from a single sentence

WorldForge generates terrain, environments and NPCs from a prompt, live in the browser.

A team selfie at a hackathon

05

Cloud & DevOps

The unglamorous half of shipping

EC2, S3, Lambda, CloudFront, Nginx, PM2, Docker and GitHub Actions, wired up and kept running.

Ujesh speaking into a microphone
The team coding side by side
Holding the cheque at the Cardano hackathon

What I build.

See every skill

The stack

Tools I reach for.

Full résumé on LinkedIn

Languages

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

Frontend

  • React
  • Next.js
  • Redux
  • Tailwind CSS
  • Three.js
  • WebGL

Backend

  • Node.js
  • Express
  • REST
  • WebSockets
  • Microservices
  • JWT auth

AI / ML

  • RAG
  • LangChain
  • Embeddings
  • Semantic search
  • Computer vision
  • Ollama

Data

  • PostgreSQL
  • MongoDB
  • MySQL
  • Prisma
  • Firebase
  • Qdrant
  • Supabase

Cloud

  • AWS EC2
  • S3
  • Lambda
  • CloudFront
  • Docker
  • Nginx
  • GitHub Actions

Projects

Built, shipped,and sometimes won.

Hackathon builds that took first place, research on its way to a conference stage, and the side projects in between. Live links where they're live, code where it's public.

  1. Hackathon buildInception hackathon · Winner · 2026

    WorldForgeBuild a world from a sentence, then train robots in it

    Won India's first world-model hackathon with a browser engine that turns a prompt into a world you can fly through. It has since grown into WorldForge: navigable AI worlds where vision-language models read the frames and RL agents train in simulation.

    • React
    • Three.js
    • TypeScript
    • VLMs
    • RL
    Live site Code

  2. ResearchPaper accepted · IMPACT-2027 · Jan 2027

    GradMeshEvery idle GPU on the network, one training cluster

    A coordinator-worker architecture that pools consumer GPUs across a LAN into one training cluster. It measures what each device can actually do, sizes the work to match and trains one shared model. Accepted at IMPACT-2027.

    • Python
    • FastAPI
    • PyTorch
    • FedAvg
    Code

  3. Hackathon buildHacker House Goa · 2026

    TidewatchFraud alerts, investigated in under a second

    Agentic fraud investigation on TigerGraph Savanna. Each alert is worked the way an analyst would, graph queries first, with no model in the decision loop: twenty alerts investigated in 0.7 seconds.

    • Python
    • TigerGraph
    • Agents
    Live site Code

  4. Hackathon buildSmart India Hackathon 2026 · PS 26121

    NWISWhat the neighbouring wells already learned

    Nearby Wells Intelligence System: decision support for drilling crews that surfaces the risks recorded in offset wells, such as mud loss and stuck pipe, along the path of the well being drilled. Built for Smart India Hackathon 2026 and demoed on synthetic data.

    • TypeScript
    • Next.js
    • AI search
    Live site Code

  5. Hackathon buildAccessibility hackathon · 2025

    BlinkyAttention that remembers on your behalf

    Holds the one thing you sat down to do, notices when you drift and quotes your own intent back to you. Frames are analysed and thrown away: the table has no image column, so privacy is enforced by the schema.

    • Next.js
    • Vision models
    • Supabase
    Live site Code

  6. Hackathon buildDenova blockchain hackathon · Winner · Solana track

    GitPayOpen-source bounties that pay out on merge

    A decentralised bounty platform for open-source work: Solidity escrow holds the reward, GitHub workflows watch the pull request, and the payout happens on its own when it merges. Won the Solana track at Denova.

    • Solidity
    • Solana
    • GitHub webhooks
    • React
    Code

  7. Hackathon buildBNB Chain hackathon · 2025

    Agents PlaygroundAI trading agents, scored on-chain

    Three AI trading agents (arbitrage, mean reversion and momentum) compete under identical market conditions on BNB Chain, with every result recorded on-chain so the leaderboard can't be argued with.

    • Node.js
    • Solidity
    • ethers.js
    • BNB Chain
    Live site Code

  8. Hackathon buildCypher 3 hackathon

    AI learning assistantA tutor with a face, grounded in the syllabus

    Personalised lessons from a realistic AI avatar, with a RAG pipeline keeping answers on the syllabus, Manim drawing diagrams on demand and eye tracking adapting the pace to the learner's focus.

    • RAG
    • Manim
    • HeyGen
    • Eye tracking
    Ask me about it
  9. The Chitoor team at work: laptops, aerial imagery and hand-traced field maps

    ResearchField survey · 2024

    Chitoor2,000 acres, mapped by drone and segmented by our own model

    A drone survey of about 2,000 acres of farmland by a team of four students. We traced plant boundaries in the aerial imagery by hand first, then trained a segmentation model to do it automatically.

    • Python
    • Drone survey
    • Image segmentation
    Ask me about it

  10. Side projectVirtual tour · college project

    Atria 360° tourThe campus, explorable from a browser

    A 360° virtual tour of the Atria campus: walk between panoramas, open hotspots and find your way around before you ever visit.

    • JavaScript
    • 3Sixty
    • WebGL
    Live site Code

  11. Hackathon buildPAC Hack · Finalist

    Form checkPose estimation that grades your workout

    Real-time pose tracking with MediaPipe and a temporal read of joint angles, so drifting exercise form is caught as it happens. A finalist at PAC Hack, Presidency University.

    • Python
    • MediaPipe
    • OpenCV
    Ask me about it

  12. Side projectApplied ML

    Skin cancer predictionCNN classifiers for skin-lesion images

    Deep-learning models trained on HAM10000 with careful preprocessing, augmentation and class balancing, judged on precision, recall and F1 for assisted screening.

    • Python
    • TensorFlow
    • CNNs
    Code
65+ more on GitHub

Internships

Three internships, real people on the other end.

A healthcare startup, an industry research collaboration with Hitachi, and an AICTE programme with Microsoft and SAP.

  1. 01

    EaseMed

    Full Stack Developer Intern

    Jan – Jun 2026 · Bengaluru

    Sole engineer on outbreak prediction.

    • Built and shipped a disease-outbreak prediction model that correlates global trade data with WHO and government health APIs to set inventory-risk guardrails.
    • A serverless ingestion pipeline in Node.js and Supabase with a Hugging Face-hosted parser, validated with 10–20 concurrent users.
    • WebSocket messaging between hospitals and vendors, a React and Next.js frontend, auth and role-based access, with inference services on EC2, S3, Nginx, PM2 and Vercel.

    The public demo is offline for now.

    The EaseMed landing page: where healthcare demand meets global supply
  2. 02

    Hitachi

    Machine Learning Intern

    2025 · 3 months · Bengaluru · with Atria

    From camera to classifier.

    • Led a student team building a ~600-image dataset for bird-species detection, from raw camera footage captured around Bengaluru.
    • Labelled it in Label Studio, then trained CNN classifiers and YOLOv11 detectors, reaching about 0.70 mAP.
    • Owned model development and pipeline design end to end, working with Hitachi researchers and academic mentors.
    A bird photographed for the species-detection dataset
  3. 03

    AICTE × Microsoft & SAP

    Machine Learning Intern

    2024 · India

    Small models that keep their accuracy.

    • An end-to-end image-classification project on real-world data: experiments, evaluation and a written technical report.
    • Compared MobileNetV2 with ResNet-50; MobileNetV2 won the efficiency-accuracy trade-off for lightweight deployment.
    • Up to 0.96 accuracy on optimised runs, with reproducible code on GitHub.
    The code

Freelance & clients

Products and sites for people who use them.

Startups and small businesses: an AI tutor in production, a wellbeing platform, a resort and a merch store, each one designed, built and shipped end to end.

  1. The Edactly home page

    Edactly

    An AI tutor that has read the whole syllabus

    Prompt in, lesson video out, for students in grades 1–12.

    • An AI pipeline in Python, Manim and Matplotlib that turns a natural-language prompt into an explainer video, grounded by RAG over the NCERT curriculum.
    • Deployed on AWS S3 and Lambda: a three-minute video in about five minutes, for roughly ₹7 of compute.
    • Authentication, user management and the APIs behind the platform's dashboards and content workflows.

    Freelance full-stack developer

    2026 – now

    Visit the site
  2. The NEXR site: 'Wellbeing, reimagined'

    NEXR

    Workplace wellbeing, without the stigma

    NEXR helps organisations remove the invisible barriers that stop employees from asking for support. The site walks visitors through it with a 3D guide.

    Website, design and build

    2026

    Visit the site
  3. The Radiant Resort home page at dusk

    Radiant Resort

    A calm address on Bannerghatta Road

    Timber chalets, Ayurvedic wellness and dining on the southern edge of Bengaluru, with a slow, cinematic site to match and booking enquiries built in.

    Website, design and build

    2026

    Visit the site
  4. The Print Perfect storefront

    Print Perfect

    Branded merch teams actually keep

    A Bengaluru merch company's storefront: a catalogue of 120-plus products, bulk ordering, and a little guide that walks first-time buyers through an order.

    Storefront, design and build

    2026

    Visit the site

  5. DRYP

    A storefront, end to end

    The full product, cart and checkout flow for an online store.

    The live store is offline; the code is public.

    Commerce build

    2025

    See the code

Seven wins, fifty-plus hackathons,and a habit of shipping by Monday.

( The short version )

I like the part of engineering where an idea has to survive real users. At EaseMed that meant hospital and vendor teams messaging in real time. At Edactly it means students watching lessons generated from a single prompt.

Outside client work I research distributed training on consumer GPUs, lead a 200-plus member code club at Atria Institute of Technology, and organise hackathons almost as often as I enter them.

B.E. Computer Science and EngineeringAtria Institute of Technology (VTU), Bengaluru · 2023 – present
Full résumé on LinkedIn
Ujesh speaking at a podium

Leadership & community

Running the room, not just sitting in it.

A code club that runs like a small company, a tech fest with a thousand applicants, and hackathons organised almost as often as entered.

Ujesh speaking into a microphone, his sticker-covered laptop beside him
A big group photo on a staircase after an event

  • Technical lead · 2023 – now

    200+members

    Code Club, Atria Institute of Technology

    Code reviews, design sessions and mentoring from idea to deployment, plus real-time tools for college departments, including a live IPL-auction simulator.

  • Tech fest organiser · 2024 – now

    1,000+applicants

    Vigyan Rang

    A ₹5L tech fest with QR-based registration and shortlisting, and five hackathons of 500-plus participants each.

  • AI lead · 2023 – 2024

    Google Developer Group On Campus

    Led the chapter's AI initiatives: technical sessions and project-based learning for students getting into applied ML.

  • Open-source contributor · Oct 2025

    7PRs merged

    LitmusChaos, Hacktoberfest 2025

    Seven accepted pull requests to the CNCF chaos-engineering project, in Go: Kubernetes resource handling and Helm chart logic.

Presenting at the Denova hackathon

Winner · Solana track · International blockchain hackathon, with GitPay.Denova

Wins

Fifty-plus hackathons. Seven wins. A different problem every weekend.

World models, blockchain, accessibility, healthcare: the theme changes every time. The habit of shipping something that works by the final demo doesn't.

Hackathon wins
7×Hackathon wins
National-level finals
20×National-level finals
Hackathons entered
50+Hackathons entered
Public projects
65+Public projects
CodeChef rating
1700CodeChef rating

On the floor

Cheques, certificates and very little sleep.

Hackathons, stages, teammates and friends: the people and the moments behind the work.

  • The team on stage with certificates at Cypher
    Cypher 1Winner
  • Presenting the build at Denova
    DenovaWinner · Solana track
  • Holding the cheque at the Cardano hackathon
    Cardano hackathonWinner
  • The team holding the Inception winner's cheque
    InceptionWinner
  • The team at the Google Solution Challenge bootcamp
    Google Solution ChallengeRegional qualifier
  • A packed hackathon floor under a 'Let the hacking begin' banner
    Let the hacking begin
  • Ujesh speaking into a microphone
    On the mic
  • Certificates in hand after Cypher 3
    Cypher 3Winner
  • Teams at the HackerRank Orchestrate finals
    HackerRank OrchestrateBronze
  • Ujesh with the Inception winner's cheque
    InceptionWinner
  • Teammates heads down at their laptops
    Heads down
  • A team selfie at a hackathon
    Team selfie
  • Ujesh pitching with a microphone
    Pitching the build
  • Ujesh at the podium
    At the podium
  • A full hall at a hackathon
    A full hall
  • A large group photo of participants
    Everyone who showed up
  • The Cardano hackathon floor
    Cardano hackathon
  • Ujesh at his laptop, a teammate in headphones beside him
    Shipping before the deadline
  • Ujesh presenting a demo
    Demo time
  • A big group photo on a staircase
    The whole crew
  • The Inception team with their cheque
    InceptionWinner
  • Two teammates working at one desk
    Pair programming
  • The team coding side by side on a hackathon floor
    Mid-build
  • Ujesh presenting to a classroom
    Running a session
  • A pile of hackathon t-shirts, bags and badges
    The swag pile
  • Teammates at a hackathon, one looking at the camera
    The crew

Research

Papers, a book chapter and work in progress.

Most of my research asks one question: how much training can you squeeze out of the GPUs people already own?

Ujesh at the podium presenting model-training results to a room
Presenting training results

Around the internet

Build logs, hackathon recaps and the occasional win, as they happen.

Off the clock

When I'm not shipping.

  • MarathonsLong runs are where the hard bugs get solved.
  • Gym, dailyNon-negotiable, deadline week included.
  • FootballWhenever there are enough people for a side.
  • DanceThe one thing on this page with no deploy step.
  • GamingArcade cabinets, shooters, anything with good movement.

Questions?Answeredright here.

Let's talk

Email me

  • Software engineering across full-stack product and applied AI or ML systems. I'm at my best on small teams where I can own a feature from schema to deploy and talk to the people using it.

  • TypeScript with React and Next.js on the front, Node.js and Express with PostgreSQL or MongoDB on the back, Python for ML, and AWS (EC2, S3, Lambda, CloudFront) to ship it. For AI work: RAG, LangChain, embeddings and vector databases like Qdrant.

  • Bengaluru, India. My Edactly work is fully remote, and EaseMed was in person in Bengaluru, so either works.

  • Cross-vendor GPU orchestration: one layer that trains across NVIDIA, Intel and Apple Silicon GPUs together. It extends GradMesh, which I'm presenting at IMPACT-2027 in January.

  • Most of it. There are 65+ public projects on GitHub. Client work stays private, but I'm happy to walk through the architecture on a call.

  • My stand-in until I can say hello myself. The model is "Robot Playground" by Hadrien59 (CC BY 4.0); the jumping, waving and section tours are wired up in code on this site.