// system.init 7th of 70,000 · 4 papers · 12+ products

Hello, I'm Ujjwal Pardeshi

ML × Full-Stack Engineer

I bridge ML/RL research and full-stack engineering — training fraud-fighting agents, shipping payment infrastructure, and publishing along the way.

// section.about

Who I Am

I'm a Computer Science undergraduate at SRM Institute of Science and Technology (B.Tech, IoT specialization, 2022–2026) with papers under review at IEEE and CVIP, and a top-10 finish among 70,000 entrants at the Meta PyTorch OpenEnv Hackathon.

My work sits where research meets shipping: GRPO-trained LoRA adapters for fraud detection, computer-vision pipelines for underwater perception, and the unglamorous plumbing — payments, caching, sync — that makes products actually run. Currently a Platform Engineer at Riamona, previously at Samsung Research Institute, Bangalore.

Ujjwal Pardeshi
// mumbai, india — utc+5:30
0 hackathon entrants ranked 7th — Meta PyTorch OpenEnv 2026
0% model accuracy emoji sentiment @ Samsung R&D
0+ variants automated Shopify gold-price engine
0 research papers 1 published · 3 under review
// section.experience

Where I've Worked

jan 2026 — present platform engineer

Riamona · mumbai

  • Architected a microservices backend — independently deployable services for fault isolation and clean scaling.
  • Ship full-stack features end-to-end in Next.js + TypeScript on Supabase (Postgres, auth, storage).
  • Integrated Razorpay payments: order creation, signature-based verification, and webhook handling.
  • Built a Shopify integration automating product, order, and inventory sync; Redis caching cut database load on high-traffic endpoints.
next.jstypescriptsupabaserazorpayshopifyredis
jan — jun 2025 ai research intern

Samsung Research Institute · bangalore

  • Researched emoji-based sentiment analysis; engineered a custom dataset of 2,700+ emoji–sentiment pairs.
  • Benchmarked neural networks, random forest, LightGBM, and ensembles over BERT + Emoji2Vec embeddings — 86.18% accuracy.
  • Findings published at IEEE CONECCT 2025, Bengaluru.
bertemoji2veclightgbmnlp
aug 2024 — jan 2025 undergraduate researcher

SRM Institute · chennai

  • Built a smart-lift occupancy detection system — OpenVINO models on a Raspberry Pi edge server, Firebase realtime sync, Next.js visualization.
  • Secured ₹80K in project funding; won "Best Innovative Solution" at AIOT Project Expo 2024.
openvinoraspberry pifirebasenext.js
// section.work

Featured Work

2026 creator & rl engineer

Chakravyuh

7th of 70,000 — Meta PyTorch OpenEnv Hackathon · $2,000 prize

A 5-agent reinforcement-learning environment for Indian UPI fraud detection. GRPO-trained two LoRA adapters — a Qwen2.5-7B analyzer and a Qwen2.5-0.5B scammer — that spar inside the environment. Live on Hugging Face Spaces.

pytorchopenenvtrl/grpolorafastapi
View on GitHub
2025 solo engineer · proprietary

Shopify Gold-Price Automation

End-to-end pricing engine for a luxury jewellery store — ~939 products, 55,000+ variants. Scrapes live IBJA gold rates twice daily, recalculates every variant with a component-wise formula, and pushes updates through Shopify's GraphQL bulk operations. Zero manual steps.

flaskpostgresqlshopify graphqlrailway
client work — code private
2025 full-stack engineer · proprietary

Invoice & Payment Portal

Enterprise platform for invoice tracking, payment scheduling, and UTR verification — slab-based early-payment discount engine, daily interest penalties, bulk Excel import, and role-based access control.

reacttypescriptexpressprismapostgresql
client work — code private
2024 researcher & developer

Underwater Reflection Removal & Depth Estimation

Presented at AI FOR OCEANS — national research conference

Hybrid deep-learning and sensor-fusion pipeline — YOLOv8, MiDaS, and OpenCV — for underwater reflection removal, object detection, and depth-based localization.

yolov8midasopencvpytorch
View on GitHub
// section.research

Research

under review

GENOS: Gaussian Estimation of Normality for Outlier Scoring in Few-Shot CLIP-Based Anomaly Detection

ujjwal pardeshi, sharanya katna — submitted to cvip 2026

under review

PINNTO: Physics-Informed Neural Networks for Trading Options

ujjwal pardeshi, mehul ashra — b.tech major project

under review

Evaluating Diverse Machine Learning Techniques for Autism Spectrum Disorder Detection: A Comparative Analysis

with the ieee srm r&d team — journal submission

// section.proof

Recognition

[01]

7th of 70,000 — Meta PyTorch OpenEnv Hackathon

Top-10 finish among 70,000 registrations with Chakravyuh, winning a $2,000 cash prize. 2026.

[02]

Best Innovative Solution — AIOT Project Expo

Won the top award and secured ₹80,000 in project funding for the smart-lift occupancy system. 2024.

[03]

Presented at IEEE CONECCT 2025

Presented published research at the IEEE Bangalore Section conference, Bengaluru. 2025.

[04]

SRMIST Director's Award

Awarded by the Directors of SRMIST for exemplary leadership at Astrophilia. Also led a 30-student R&D team at IEEE SRM. 2024.

// section.stack

Tools I Reach For

languages
python · typescript · javascript · sql
ml / cv
pytorch · tensorflow · scikit-learn · lightgbm · hugging face · langchain · opencv · yolov8 · openvino · bert · lora · grpo · openenv
full-stack
next.js · react · node.js · fastapi · rest apis · jwt
data
postgresql (supabase) · redis · firebase · chromadb
cloud / devops
aws · gcp · docker · ci/cd · git
platforms
razorpay · shopify · raspberry pi · nvidia a100s
11 certifications
// section.activity

GitHub Activity

contributions in the last year

less more

full profile on github →

// terminal.ask

Ask Me Anything

ujjwal@portfolio — ask me anything
// open.channel

Let's build something.

Open to roles and collaborations in ML engineering, applied research, and full-stack work.