Home

I’m Yashovardhan Srivastava (Yash is fine), a recent engineering graduate from NIT Warangal (2025). I build developer tools, research systems, and practical machine learning projects, then write about what I learn along the way, both technically and otherwise.

Reach me at ysrivastava82(at)gmail(dot)com or on Twitter.

By the numbers

  • 70+ public projects across research, developer tools, and experiments, with some featured on HN, GitHub, and Twitter. 2x Kaggle Expert with popular notebooks.
  • Worked at 3 startups and 1 research lab across AI, fintech, and infra - shipped products with real users and co-authored research at IIT-BHU.
  • I don’t rely on talent. I rely on obsession. I take ownership of what I build and go unreasonably far to make it work.

Resume if you prefer that format.

Projects 🧰

Here I’ll pin some of my favorite projects, more on the research 👨‍🔬 side(Feel free to critique me on this(and try to contribute if possible :) ) :

  • Cadence : Cadence implements an evolutionary system that uses LLMs to iteratively generate, mutate, and improve programs for solving computational problems such as the Traveling Salesman Problem (TSP).

  • REALBT : REALBT is a simple, effective backtesting engine written in pure Python.

  • Arrakis : Arrakis is a library to conduct, track, and visualize mechanistic interpretability experiments. 28+ stars on GitHub ; 250+ monthly PyPI downloads.

  • Deeprobe : Deeprobe studies feature importance and pattern understanding in Sparse Autoencoders using Monte Carlo Tree Search.

  • SAE Macaronic Languages : Understand whether language models learn words beyond language barriers, a study in mechanistic interpretability.

  • Secure BPE : Research prototype exploring a modified, security-aware Byte Pair Encoding algorithm.

  • Collaborative Debating : Experimental implementation of multi-agent debate for improving LLM reasoning and factuality.

  • NEAT-JAX : An implementation of Neuroevolution of Augmented Topologies Algorithm in JAX which is compatible with EvoJAX. 14+ GitHub stars ; multiple PRs.

  • Nexus Theory : Can we really trust our human-ness for the messages that we send into the cosmos? Nexus theory is a gamified version to understand machine learning interpretability using Large Language Models.

  • Elixr : Elixr an autograd library using Complex Numbers similar to Pytorch. 3+ Github stars ; Multiple PRs

  • Attention Free Revolution : Developed Leviathan architectures, and alternate to Transformer architecture using a modified attention scores, taking inspiration from signal processing. 7+ Github stars

  • P-GLAm : P-GLAm is a random thought experiment on Infinite Monkey Theorem. In this, I developed a GPT-2 inspired Large Language Model which aims to test the arithmetic correctness.

Here I’ll pin some of my favorite projects, more on the development 💻 side. Feedback is always appreciated for projects like these.:

  • PySlides: PySlides is terminal based application that converts markdown into slides that can be presented from the terminal.

  • Devsidian : Personal developer journal for logging your development journey, built with Lovable for personal use.

  • Snappyr : Setup Python Projects Blazingly Fast, and work on things that matter. No External Dependencies.

  • Dynamo : Dynamo is a Python/Rust implementation of a load balancer and autoscaler for MySQL web tier.

  • Safe SQL : Safe SQL provides sanity checks for common DB pitfalls(so you don’t delete prod DB) ; available as a python package(CLI included). 300+ monthly PyPI Downloads.

  • Stock Tank : End to End ML pipeline to predict stock prices(upto 30 days). Automate retraining, evals and more(Github Actions). Streamlit Web App available as well.

  • Gym Tunes : GymTunes is a simple AI agent that schedules a random playlist into your GCalendar based on your vibe.

  • AI GF : A small weekend project that through which you can create a virtual girlfriend(not made for imitation, but for learning)

  • Synapse : Synapse is a Hacker News style forum platform for communities.

  • grove : Terminal knowledge garden - Obsidian-style note-taking for the CLI, built in Go with Bubble Tea. Features wiki-links [[]], vault-wide AI search, note templates, and a full TUI with vim keybindings.

  • pairy : Neovim AI pair programmer. Uses Gemini directly inside Neovim for code completion, explanation, and refactoring - no API proxy, no bloat.

  • Pandora : Pandora is domain agnostic framework for case study generation and solving.

  • Verizon : A Git like version control system, from scratch, in Python, spelled out.

  • YeetCode : YeetCode is a sassy version of Python made for all GenZ people. The aim is to create a new programming language which is bussin’.

  • Blaze : RAG system built with Cohere, Metaphor, LangChain, and Chainlit, originally created during a recruitment process and deployed on Hugging Face. 8+ GitHub stars.

  • CodeSmith : Developed a ChatGPT-inspired chatbot trained on a Python programming problems on custom created dataset, made using Langchain, and deployed on Huggingface.

  • Alzhemist : One of the first projects that got me in to the world of Attention. A Deep Learning Model to see which classifies Brain MRI on the basis of the dementia (AD). The images are classified as follows - Mildly Demented, Moderate Demented, Non Demented, Very Mild Demented.

  • Maxwell : One of my most priced possession. Maxwell is twisted take on One Shot Frequency Dominant Neighborhood Search. The scheme provided in the paper is a bit modified to generate fingerprint for an image.

  • SpiceyDicey : SpicyDicey is a end to end machine learning project that aims to predicts the number that appears on a dice. All of the work in collecting the data and editing the images has been done individually and from scratch.

Here are some of the awesome notebooks 📓 I’ve made on Kaggle(I’m a 2x Kaggle Expert also !!) :

  • FC Barcelona is Back! : Analyzed FC Barcelona’s LaLiga performance in the 2022-23 season on Kaggle, achieving Bronze Medal and 200+ views apart from receiving recognition from Kaggle.

  • BART Pretrainig from Scratch : Developed a BART model from scratch using Huggingface on Shakespeare dataset in a notebook on Kaggle, which received a silver medal and 600+ views.

  • Tensorflow Recommendation System : Demonstrated on using Tensorflow Recommendation System in a Kaggle notebook that gained bronze medal, and 500+ views.

Experience 👷 :

  • Product Engineer, Glomopay: Part of Cards and Subscription Team, handling cross border payments and solving business critical problems daily.

  • AI Engineer Intern, TurboML: Shipped end to end self hosted video generation(700K+ impressions), image generation/editing(2.7M+ impressions), and customized meme generation to Whatsapp, utilizing state of the art diffusion models. Facilitated resolution of STT server memory outage, leading to 63% decrease in memory usage

  • DevOps Engineer, Strykr.ai: Primary Backend Engineer of Strykr.ai, where I worked on implementing request caching, response streaming, and async API call which reduced latency by 6 seconds and facilitated deployment migration of the said application from Render to Railway.

  • Project Intern, Solvendo India Private Limited: Worked with the Machine Learning Team on : a) A production LLM RAG application b) developing time series machine learning models for predict volatility of a stock(Deep Learning Based, GluonTS).

  • Research Intern, Indian Institute of Technology-Banaras Hindi University: Worked under Prof. A.K. Singh on a research project on developing a machine translation system for low resource languages such as Hindi, Bhojpuri, Magahi, Maithali.

  • President, Big Data Analytics and Consulting Cell(National Institute of Technolgy, Warangal) : Lead the BDACC team for the academic year 2024-2025 after being the member for 2 years in a team that has collaborated in several of the student club events such as Kaggle, Pytorch Workshop and Case study competitions, among other initiatives to develop a community of machine learning enthusiasts in NIT Warangal.

  • Executive Member, Research and Development Cell(National Institute of Technolgy, Warangal): Part of Undergraduate Research Association team of NIT Warangal which actively takes part in educating and fostering academic growth among undergraduate students.

About Me 🙇‍♂️:

I’m Yashovardhan Srivastava (Yash is fine), engineering graduate from NIT Warangal, 2025. I’ve been fascinated by computers for as long as I can remember, and that fascination compounded into obsession. Most of my projects started as a random thought - “I wonder if that works” or “I should just build that” - and went from there. That loop has made me a decent programmer.

I’m a researcher in my free time and an engineer full time. I come up with ideas, test them, write about them, ship them.

Career Goals 🥅:

I want to work on things that are genuinely hard and worth doing - research-quality problems with engineering-quality standards. I’m not looking for a title; I’m looking for a problem worth being obsessed about.

What I look for:

Ideas flowing freely regardless of where they come from, a clear mission to orient work around, and a team good enough that I’m learning from them. I’m early in my career and care more about the environment than the perks.

Proficiency and Interests ⭐ :

  • Research Interests: Mechanistic interpretability, NLP, neuroevolution, systems for ML
  • Languages: Python, Ruby, C/C++, Lua, Go (picking it up)
  • Frameworks: PyTorch, JAX, TensorFlow, Ruby on Rails, Keras
  • Personal Interests: Football, badminton, reading (a lot), writing

Achievements 🥇

  • President, Big Data Analytics and Consulting Cell, NIT Warangal (2024-2025)
  • 2x Kaggle Expert - Notebook rank 699, Dataset rank 573

Research Papers I Love 📎

In no particular order, I am listing some really awesome research papers that in one way or other have helped me think outside of the box.

If something here caught your eye - a project, a blog post, an idea - feel free to reach out. ysrivastava82@gmail.com or Twitter. I reply.