Turning data into decisions.

I'm Sidharth Choudhary, a final-year dual-degree student in Data and Computational Science at IIT Jodhpur. I build machine learning and AI systems end to end — from the math underneath to the code that ships. My work spans data science, ML engineering, and the newer wave of agentic AI, with a particular fondness for problems where finance and markets make the stakes real.

Built to ship, not just to demo.

Most student projects die in a notebook. Mine don't. I care about the full path — clean data, models that earn their complexity, and systems that actually run when someone other than me uses them.

The problems range from volatility forecasting and risk modeling to language translation and agentic workflows. Different domains, same standard: get the math right, build it properly, and make it hold up under real load.

Strategy vs buy-and-hold equity curve for a Nifty 50 quant strategy

The tools I work with.

Core

· Python

· SQL

· NumPy

· pandas

ML & DL

· scikit-learn

· PyTorch

· TensorFlow

AI & Agents

· LLMs

· Agents

· RAG

Quant

· GARCH

· statsmodels

· Time-series

Deployment

· PostgreSQL

· Streamlit

· FastAPI

· Git

Production-grade ML projects
5

Optimizers rigorously compared
4

Best BLEU score (EN-HI NMT)
18.22

Volatility models benchmarked
6

How I actually work.

I start with the math — understanding the problem deeply before writing a line of code. Then I build: clean pipelines, models that justify their complexity, and validation that holds up to scrutiny. No shortcuts, no black boxes I can't explain.

The goal is always the same — work that survives contact with real data and real users, not just a clean number on a slide.

Baseline vs regime-aware portfolio equity curve

Selected Projects

Quant Volatility & Risk Engine

Volatility & risk system for Nifty 50 using GARCH-family models — regime detection, stress testing, and regime-aware allocation.

Real-Time Analytics & Insight Engine

Deployment-grade pipeline on UCI Online Retail II — cleaning, analysis, and an automated insight layer, served live via Streamlit + FastAPI.

SQL Analytics Dashboard

Industry-grade dashboard on a real relational dataset — advanced PostgreSQL: window functions, CTEs, and analyst-style query patterns.

Comparative Backprop for NMT

Optimizer comparison (SGD, Adam, RMSProp, Adagrad) on an English–Hindi Seq2Seq + LSTM + attention model. Best BLEU: 18.22 (RMSProp).