Skip to content
View anantshri1's full-sized avatar
6 followers · 10 following

Block or report anantshri1

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 is supported. This note will only be visible to you.
Report abuse

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

Report abuse
anantshri1/README.md

Hi there!

Hi, I am Anant, a theoretical physicist about to finish a PhD in Particle Physics, with a focus on strongly-interacting low-dimensional quantum systems. Welcome to my GitHub page!

  • 🔭 I’m currently wrapping up my PhD on three-dimensional mirror duality, which is, at its heart, a very interesting (read: weird) phenomenon in three spacetime dimensions that relates multiple field theories.

My research focuses on classifying dual pairs in three-dimensional supersymmetric quantum field theories with minimal supersymmetry. This has been an open problem for three decades — the absence of extended supersymmetry, in conjunction with the presence of topological interactions, removes the usual geometric constraints from string theory that make such classifications tractable. My doctoral work developed a systematic framework for constructing these dual pairs: starting from known dualities as seeds and applying controlled operator deformations to generate new ones. This produced the first comprehensive results in this regime, with findings published in Physical Review D and the Journal of High Energy Physics.

📄 Find my publications here.

  • 🌱 Alongside the PhD, I've been building ML/AI engineering skills from the ground up. My scientific programming language of choice is Python (NumPy, pandas, matplotlib, Seaborn), but I am familiar with MATLAB and Mathematica.

I have hands-on experience across the ML stack:

  • Classical Regression and Classification via scikit-learn, XGBoost, CatBoost, and imblearn
  • Deep Learning via TensorFlow/Keras (FFNs, CNNs), and PyTorch
  • Geometric Deep Learning via PyTorch Geometric
  • Domain-adapted LLM fine-tuning
  • Retrieval-Augmented Generation (vector- and graph- based) mostly via LlamaIndex
  • NLP and Transformer architectures via nltk, spaCy, HFTransformers
  • Time-series forecasting via LSTMs, multi-head self-attention Transformers and neuralforecast
  • Tracking and logging ML experiments via Mlflow
  • PPO Reinforcement Learning Implementations
  • Experiment-driven model development and evaluation

📍 Currently: Learning SQL and LangChain based agentic workflows by making an nl2sql pipeline.

The repos below are the output of that. (I am currently fighting Google Colab environment dependencies as I attempt to learn MLOps workflows - Colab is winning at the time of writing)

Pinned Loading

  1. gdl_orbitwars gdl_orbitwars Public

    A collection of increasingly sophisticated agent architectures designed to compete in Orbit Wars, culminating in Proximal Policy Optimization (PPO) via Graph Neural Networks (GNNs).

    Python

  2. GraphRAG_3dQFT GraphRAG_3dQFT Public

    Hybrid Graph+Vector RAG pipeline built over complex results on three-dimensional quantum field theories. Extends previous RAG pipeline. Includes a PoC trained on a single paper, and a full pipeline…

    Jupyter Notebook

  3. domain_adapted-LLM_fine-tuning domain_adapted-LLM_fine-tuning Public

    Domain adaptation and fine-tuning of Qwen2.5 3B Instruct on scientific text.

    Jupyter Notebook

  4. RAG_3dQFT RAG_3dQFT Public

    RAG pipeline built on Mistral-7B over complex results on three-dimensional quantum field theories. Compares performance of Vanilla RAG and Hierarchical RAG to assess quality of retrieval.

    Jupyter Notebook

  5. timeseries_forecasting_btc_prediction timeseries_forecasting_btc_prediction Public

    Benchmarking a stacked LSTM, custom Transformer, and a Temporal Fusion Transformer on forecasting tasks using hourly Bitcoin data.

    Jupyter Notebook

  6. nl2sql nl2sql Public

    End-to-end NL→SQL pipeline using LLMs to generate, execute, and refine SQL queries from natural language questions. Features schema-aware prompting, self-correction loops based on execution feedbac…