Skip to content
View Mithil-hub's full-sized avatar

Block or report Mithil-hub

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don't include any personal information such as legal names or email addresses. Markdown supported. This note will be visible to only you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Mithil-hub/README.md

Hey, Mithilesh Kothandaraman Here!!

AI Researcher & Engineer — MS Data Science @ Arizona State University
Specializing in LLMs · Generative AI · CAD Multimodal Systems · Market Microstructure

LinkedIn Email Profile Views


About Me

I build AI systems that sit at the intersection of research and production — from training transformer-based models and fine-tuning LLMs to designing multi-agent pipelines and analyzing high-frequency trading data.

  • 🔬 Research: Spatial-aware CAD geometry-aware AI, transformer policy models, multimodal diffusion
  • 🤖 LLMs & GenAI: RAG pipelines, LangChain/LangGraph agents, LoRA/PEFT fine-tuning, CAD geometry-aware generation, MCP server integration
  • 📈 Quant Finance: High-frequency trading data analysis, order flow imbalance, limit order book reconstruction, market microstructure
  • 🎓 Education: MS Data Science, ASU (GPA 3.78) · BS Computer Science, SRM University (GPA 3.8)
  • 🌍 Location: Tempe, AZ · Open to 2026 New Grad & Internship roles ( Relocation within USA )

Featured Projects

🧠 LLM & Generative AI

Project Description Stack
MoE Stable Diffusion Mixture-of-Experts vs Dense baseline for image generation — 30% compute reduction, 20% faster inference PyTorch · Diffusers · CLIP · T5
STaR Reasoning Self-Taught Reasoner on GSM8K — Zero-Shot 47% → STaR 68% exact match with Llama 3.2-3B PyTorch · HuggingFace · LangChain
AlphaLoop Multi-agent LLM trading system with RL reward modeling, episodic trade memory, LoRA fine-tuning LangGraph · PEFT · FastAPI · SQLite

📊 Quantitative Finance

Project Description
HFT-Market Microstructure Analytics 🔒 Reconstructed a live limit order book from 50,806 nanosecond-precision MBO events using vectorized q queries — OFI signals, 27 microstructure features, market making & stat arb · Private repo — message me on LinkedIn for access

Learning & Exploration

Quick implementations built to deepen understanding of specific AI techniques.

Topic What I Built Stack Repo
Email RAG Privacy-preserving personal email Q&A — Gmail OAuth, pgvector semantic search, multi-user isolation LangChain · pgvector · Mistral · Ollama View
AI Travel Planner LangGraph ReAct agent — weather, places, currency and expense tools with FastAPI + Streamlit LangGraph · FastAPI · Streamlit · Groq View
RecSys Benchmark MF vs NCF vs DMF on MovieLens — RMSE, Precision, Recall comparison Keras · TensorFlow · MovieLens View

Technical Stack

AI & ML

LLMs · Transformers · RAG · LoRA/PEFT · Diffusion Models · Reinforcement Learning
Mixture-of-Experts · Multimodal AI · Fine-tuning · Model Optimization · Causal Analysis

Frameworks & Tools

PyTorch · TensorFlow · HuggingFace · LangChain · LangGraph · RLlib · OpenAI Gym
FastAPI · MLflow · Airflow · Apache Spark · Databricks · Scikit-learn

Programming

Python · KDB+/Q · SQL · R · C/C++ · Java · Bash

Cloud & Infra

AWS SageMaker · Google Cloud AI · Microsoft Azure · Docker · Git · MLOps

Research

Arizona State University — Graduate Research Assistant (March 2025 – Present)

  • Designing spatial-aware CAD generation models for geometric reasoning
  • Built a 14.5M parameter transformer-based policy model for sequential parametric CAD geometry → 74% improvement in reconstruction accuracy
  • Engineered image reconstruction pipelines processing 10,000+ 3D parametric models
  • Optimized data workflows across 15M+ record datasets

Currently Working On

  • 🔧 Extending STaR bootstrapping to multi-step tool-use reasoning in LLM agents
  • 📐 Spatial reasoning enhancements for transformer-based CAD policy models
  • 📊 Adding PPO/DQN policy training loop to AlphaLoop multi-agent system

Open to research collaborations, internships, and new grad roles in AI/ML, Data Science, and Quantitative Finance.

Pinned Loading

  1. Optimizing-Multimodal-Diffusion-Transformers-with-MoE-Enhanced-Stable-Diffusion-3 Optimizing-Multimodal-Diffusion-Transformers-with-MoE-Enhanced-Stable-Diffusion-3 Public

    Mixture-of-Experts vs Dense baseline for InstructPix2Pix image editing using Stable Diffusion 1.5

    Python

  2. STaR-Self-Taught-Reasoner-Reasoning-Enhancement-for-LLMs-on-GSM8K STaR-Self-Taught-Reasoner-Reasoning-Enhancement-for-LLMs-on-GSM8K Public

    STaR Self-Taught Reasoner implementation on GSM8K — Zero-Shot CoT vs Vanilla SFT vs STaR with Llama 3.2-3B

    Python

  3. AlphaLoop-Self-Improving-Multi-Agent-Trading-System-with-RL-Feedback AlphaLoop-Self-Improving-Multi-Agent-Trading-System-with-RL-Feedback Public

    RL reward modeling + episodic trade memory + LoRA fine-tuning pipeline built on top of a multi-agent LLM trading system — LangGraph, LangChain, PEFT

    Shell