Resume

Mon May 25 2026

Technical Skills

  • Languages: Python, C++, JavaScript
  • Libraries: numpy, pandas, matplotlib, sympy
  • Framework: PyTorch, FastAPI, Qdrant, Airflow, Docker
  • Numerical Analysis
  • Mathematica
  • LaTeX

AI/ML - All the codes are available on my GitHub profile

  • Built a transformer model from the paper Simplifying Polylogarithms with Machine Learning to simplify polylog expressions. Devised an architecture to use transformer model in the RL to make the simplification interpretable.
  • Built a data pipeline to fetch flight data from cloud and weather data from API, process it and merge them.
  • Built a RAG tool to answer user query from the document they uploaded.
  • Used the Gemini API to categorize articles from a webpage and did data processing to format the articles into a list in markdown format.
  • Wrote a decoder-only transformer from scratch.
  • Built denoising diffusion model.
  • Demonstrated superposition in a neural network from (Toy Models of Superposition) (partial)
  • Built RNN and LSTM and tested on the copying task.
  • Reproduced Discovering Physical Concepts with Neural Networks for three problems, demonstrating the network’s ability to learn physical quantities from observation.
  • Wrote an autoencoder using Tensorflow and tested on MNIST dataset.
  • Developed a neural network in Python from scratch.
  • Hands-on experience on Machine learning projects.
    • Used regression technique, cross-fold validation, and PCA to study texture data.
    • Used regression, decision tree, and random forest to study and compare the wine quality data.

Mathematica

  • Developed a Mathematica code for one-loop and two-loop calculation of an observable in the KMOC formalism.
    • The code implements custom derivatives and various simplification methods to simplify and manipulate expressions.

Numerical Analysis

  • Metropolis algorithm for sampling from a distribution.
  • Studied phase transition in 2D Ising model and calculated critical scaling exponents from scaling relations.
  • Differential Equation Solvers: Advection, Hyperbolic, and Parabolic. Used Crank-Nicholson scheme, Runge-Kutta, and velocity Verlet methods.
    • Astrophysics - Studied relations between mass, density, and radius of a neutron star using the Bethe-Jhonson model for the Equation of State.
    • Stochastic Modelling - Solved for the steady state protein distribution for two and three stages models for Stochastic Gene Expression.
  • Integration techniques: Simpsons, Trapezoid, Bode rule, Monte-Carlo method. Used GSL routine for Gaussian Quadrature.
  • Interpolation techniques: Lagrange’s method, Neville algorithm.

Papers

  • Co-authored Log-soft constraints on KMOC - 2026

Achievements

  • Joint Entrance Screening Test (JEST) - 2022 National qualifying exam for physics graduate programs at research institutes in India. Rank: 49
  • Joint Admission Test for Masters (JAM), Physics - 2022 National qualifying exam for admission in IITs for graduate programs. Rank: 267
  • Lakshyodya - 2018 Inter-school C++ programming competition. Rank: 3

[Your Full Name]

[Your City, State, Zip] | [Your Phone Number] | [Your Email Address] GitHub: github.com/avsh12[cite: 1] | LinkedIn: [Your LinkedIn URL] | Portfolio/Website: [Your Website URL]

Summary

Technical researcher with a strong foundation in physics and mathematics, seeking to transition into empirical AI research within the Anthropic Fellows Program[cite: 2]. Fluent in Python with extensive experience building ML systems, training neural networks from scratch, and conducting complex numerical analysis[cite: 1, 2]. Specifically focused on reducing risks from advanced AI systems through mechanistic interpretability, while bringing practical engineering rigor to ML systems performance and reinforcement learning environments[cite: 2].

Education

  • [Degree, e.g., Ph.D. in Theoretical Physics] | [University Name], [Location] | [Expected Graduation Year]
  • [Degree, e.g., M.Sc. in Physics] | [University Name], [Location] | [Graduation Year]
  • [Degree, e.g., B.Sc. in Physics] | [University Name], [Location] | [Graduation Year]

Technical Skills

  • Programming Languages: Python, C++, JavaScript[cite: 1]
  • Machine Learning & Data: PyTorch, TensorFlow, FastAPI, Qdrant, Airflow, Docker, numpy, pandas, matplotlib, sympy[cite: 1]
  • Mathematical & Analytical Tools: Mathematica, Numerical Analysis, LaTeX[cite: 1]

Research & Engineering Experience

AI Safety & Mechanistic Interpretability

  • Demonstrated superposition within a neural network, applying concepts from the Toy Models of Superposition framework (partial implementation)[cite: 1].
  • Devised an architecture integrating a transformer model within a Reinforcement Learning setup to make the simplification of mathematical expressions interpretable[cite: 1].
  • Constructed a transformer model based on the paper Simplifying Polylogarithms with Machine Learning to accurately simplify polylogarithmic expressions[cite: 1].

ML Systems, Data Pipelines & Performance

  • Architected and built a comprehensive data pipeline to extract flight data from cloud infrastructure and integrate it with API-sourced weather data[cite: 1].
  • Developed a Retrieval-Augmented Generation (RAG) tool capable of processing user uploads to answer query-based document questions[cite: 1].
  • Utilized the Gemini API to categorize web articles, implementing data processing logic to format the output into structured markdown lists[cite: 1].

Deep Learning & Reinforcement Learning

  • Wrote a decoder-only transformer and a standard neural network entirely from scratch in Python to deepen understanding of model internals[cite: 1].
  • Built and trained a denoising diffusion model[cite: 1].
  • Developed RNN and LSTM models, successfully testing their capabilities on the copying task[cite: 1].
  • Reproduced the results of Discovering Physical Concepts with Neural Networks across three distinct problems, proving the network’s ability to learn physical quantities directly from observational data[cite: 1].
  • Developed a TensorFlow-based autoencoder and evaluated its performance utilizing the MNIST dataset[cite: 1].
  • Applied regression techniques, cross-fold validation, and PCA to analyze texture data[cite: 1].
  • Utilized regression, decision trees, and random forest models to analyze and compare wine quality datasets[cite: 1].

Applied Physics & Numerical Analysis

  • Developed custom Mathematica code for one-loop and two-loop calculations of observables within the KMOC formalism, implementing custom derivatives and simplification techniques[cite: 1].
  • Utilized the Metropolis algorithm to sample distributions and studied phase transitions in the 2D Ising model to calculate critical scaling exponents from scaling relations[cite: 1].
  • Implemented Crank-Nicholson, Runge-Kutta, and velocity Verlet methods to solve advection, hyperbolic, and parabolic differential equations[cite: 1].
  • Investigated the relationships between mass, density, and radius of neutron stars using the Bethe-Johnson model for the Equation of State[cite: 1].
  • Calculated steady-state protein distributions for two- and three-stage models applied to Stochastic Gene Expression[cite: 1].
  • Applied multiple integration (Simpsons, Trapezoid, Bode rule, Monte-Carlo, Gaussian Quadrature via GSL) and interpolation techniques (Lagrange’s method, Neville algorithm) for numerical modeling[cite: 1].

Publications

  • Co-authored: Log-soft constraints on KMOC (2026)[cite: 1]

Achievements

  • Joint Entrance Screening Test (JEST) - 2022: Rank 49 in the national qualifying exam for physics graduate programs at research institutes in India[cite: 1].
  • Joint Admission Test for Masters (JAM), Physics - 2022: Rank 267 in the national qualifying exam for admission to IIT graduate programs[cite: 1].
  • Lakshyodya - 2018: Rank 3 in the inter-school C++ programming competition[cite: 1].