Open to ML engineering internships

ShantanuSingh

AI/ML engineer and two-time national hackathon Grand Finalist. I build models and then build the software around them.

national Grand Finalist: 2×; national Grand Finalist
ML projects shipped: 6; ML projects shipped
pair F1 on LedgerLens: 0.999; pair F1 on LedgerLens

Grand Finalist — Meta × PyTorch × Hugging Face OpenEnvGrand Finalist — iDEA 2.0, Union Bank of IndiaSmart India Hackathon 2026Top 100 of 31,000 teamsTop 40 of 1,500 teams

I'm a computer science (AI & ML) student who ships machine learning end to end — from training and evaluating models to deployed apps people actually use. I've built explainable fraud detection, RAG pipelines, an RL environment for fine-tuning LLMs and live 3D data platforms, mostly with Python, PyTorch, FastAPI and React.

Studying: B.Tech CSE (AI & ML), JSS University, 2025–2029
Based in: Greater Noida, India
Working on: Machine learning, generative AI, full-stack apps

Selected work

Six projects, from training runs to deployed apps. The first two reached national hackathon finals.

01 / 06

OpenEnv Grand Finalist — top 100 of 31,000 teams

ML Debug Env

An RL environment that teaches LLMs to debug PyTorch code

1.5BQwen2.5 fine-tuned with GRPO + LoRA

A partially observable reinforcement-learning environment where an LLM agent diagnoses and fixes broken PyTorch training scripts through diagnostic tool calls.

Rewards only fixes that pass sandboxed subprocess grading, AST validation and an LLM judge of the diagnosis.
An adversarial task scheduler serves harder bugs as the agent improves.

PyTorch
Qwen2.5
LoRA
GRPO
RL

02 / 06

iDEA 2.0 Grand Finalist — top 40 of 1,500 teams

SentinelIQ

Real-time insider fraud detection with explained alerts

3model ensemble, SHAP on every alert

Per-employee behavioural baselines feed an Isolation Forest, LSTM autoencoder and XGBoost ensemble. Every alert is explained with SHAP and grouped into fraud cases.

Alerts are auto-grouped into fraud cases instead of a noisy feed.
Served through FastAPI so SIEM tools can consume them directly.

Isolation Forest
LSTM
XGBoost
SHAP
FastAPI

03 / 06

LedgerLens

AI copilot for GST reconciliation

0.999pair F1 vs 0.98 rule baseline

Reconciles invoices against the ledger, bank statement and GSTR-2B, and prices every mismatch in rupees.

Gradient-boosted record matchers trained on 55,000+ candidate pairs, tested on a held-out month.
Graph analysis catches PAN-linked supplier rings.
Deployed on Hugging Face Spaces and Vercel.

scikit-learn
FastAPI
Next.js
NetworkX
Docker

Code

04 / 06

CourtMitra

Indian court judgments, in plain language

RAGgrounded Q&A with hallucination guards

Turns court judgments into plain-language summaries, legal entities, reasoning flowcharts and Hindi translations.

RAG pipeline on PyMuPDF and ChromaDB for grounded questions and similar-case search.
Entity extraction with spaCy and sentence-transformer embeddings.

RAG
ChromaDB
sentence-transformers
spaCy

05 / 06

Smart India Hackathon 2026

SAMUDRA

3D ocean data explorer for INCOIS

463Argo floats in one 3D scene

An interactive 3D platform fusing NOAA WOA23 ocean fields with Argo floats, gliders, BGC floats and real HYCOM currents.

Plug-in Python ingest adapters write to an offline store under 15 MB.
Bias and RMSE validation views for comparing data sources.

Next.js
React Three Fiber
xarray
Python

Code

06 / 06

Solar System Lab

Planetary science in the browser

A*Mars EVA route planner

A 3D orrery driven by NASA JPL orbital elements, solving Kepler's equation every frame with Newton's method.

Route planner respects oxygen and radio limits.
Monte Carlo rover-survival model and a landing-site scorer.

JavaScript
Three.js
Numerical methods

Live demo

Skills, in use

Each lit cell is a skill I used to ship a project. Hover a cell to follow its row and column.

Which skills each project used
	ML Debug Env	SentinelIQ	LedgerLens	CourtMitra	SAMUDRA	Solar System Lab
Python	used	used	used	used	used
Deep learning	used	used
Classical ML		used	used
LLMs & RAG	used			used
Reinforcement learning	used
Explainable AI		used
NLP				used
Graph analysis			used
FastAPI		used	used
React / Next.js			used		used
Three.js					used	used
Numerical methods					used	used

AI and ML

Machine Learning
Deep Learning
NLP
Anomaly Detection
Explainable AI (SHAP)
Feature Engineering
PyTorch
TensorFlow
scikit-learn
XGBoost
Pandas
NumPy

Generative AI

LLMs
RAG
ChromaDB
sentence-transformers
LoRA
GRPO
Reinforcement learning
spaCy
Groq API
Gemini API

Programming

Python
C++
C
SQL
Java
JavaScript
TypeScript

Software

FastAPI
Flask
REST APIs
React
Next.js
Three.js
PostgreSQL
MongoDB
MySQL
SQLite

Tools and cloud

Git
Docker
Hugging Face Spaces
Vercel
Netlify
Render
Jupyter
Tableau

Recognition

Each square is a team. The lit ones made it to the grand final.

Grand Finalist, 2026

Meta × PyTorch × Hugging Face OpenEnv Hackathon

Top 100of 31,000 teams

One square = 100 teams

Built ML Debug Env

Grand Finalist, 2026

iDEA 2.0, Union Bank of India

Top 40of 1,500 teams

Built SentinelIQ

So far

From the first semester to the next step.

2025
Started B.Tech CSE (AI & ML)
JSS University, Noida. Coursework in machine learning, DSA, Python, C/C++ and database systems.
2026
Smart India Hackathon — SAMUDRA
A 3D ocean data explorer for INCOIS built on real NOAA, Argo and HYCOM data.
2026
Grand Finalist, iDEA 2.0
Top 40 of 1,500 teams with SentinelIQ, a real-time insider-fraud early-warning system.
2026
Grand Finalist, OpenEnv Hackathon
Top 100 of 31,000 teams with an RL environment that trains models to debug code.
2029
Graduation (expected)
Looking for ML engineering internships in the meantime.

Let's build something.

I'm looking for ML engineering internships and interesting problems. Email is the fastest way to reach me.

Greater Noida, India

Resume

ShantanuSingh

Selected workSelected work

ML Debug Env

SentinelIQ

LedgerLens

CourtMitra

SAMUDRA

Solar System Lab

Skills, in useSkills, in use

AI and ML

Generative AI

Programming

Software

Tools and cloud

RecognitionRecognition

Meta × PyTorch × Hugging Face OpenEnv Hackathon

iDEA 2.0, Union Bank of India

So farSo far

Started B.Tech CSE (AI & ML)

Smart India Hackathon — SAMUDRA

Grand Finalist, iDEA 2.0

Grand Finalist, OpenEnv Hackathon

Graduation (expected)

Let's build something.Let's build something.

Selected work

Skills, in use

Recognition

So far

Let's build something.