# llms.txt for https://adarshdubey.com # Generated: 2026-07-06 # See https://llmstxt.org/ for specification ## SITE Name: Adarsh Dubey Description: Personal portfolio and blog of a Full Stack AI Engineer Author: Adarsh Dubey URL: https://adarshdubey.com ## CONTENT Type: Technical blog posts, project documentation, devlogs, reading list, and experimental projects Topics: Machine learning, deep learning, NLP/LLMs (BPE tokenization, fine-tuning, SLMs), web development (Next.js, React, TypeScript, FastAPI, Tailwind CSS), system design, MCP (Model Context Protocol), autograd engines, tooling and infrastructure, developer productivity, GSoC, open source contributions Content style: Practical, hands-on tutorials with code examples; deep dives into ML concepts; project post-mortems and build logs; GSoC journey reports; opinion pieces Pages: 6 main sections: / (home) — landing page with header, career highlights, top 5 blog posts /projects — portfolio projects + experimental/toy projects + technologies used /blogs — 18 published articles with tag-based filtering /logs — devlogs (fetched from GitHub) /reading — curated reading list (blogs, books, resources) Blog posts cover: - MCP server for FastAPI - BPE tokenizer implementations and theory - Gemma model fine-tuning and GSoC journey (community bonding, midterm, final report) - Small language model revolution - Next.js and AutoSend newsletter app - NumPy and batch processing - Terminal workflow, UX laws, code formatting tools - College clubs, Comment Mind, Syncloop - First steps in machine learning Projects documented: - Facet: Gemma finetuner with efficiency focus - Incliarray: Custom N-dimensional array library in C++ - Anon: Privacy-first anonymous forum - Inclinet: Scratch-built neural network library - FastAPI MCP Inspect: MCP server integration for FastAPI - Inclitoken: BPE tokenizer from scratch - pplx-cli: Perplexity AI CLI tool - Various C++ ML scratch implementations (k-means, logistic regression, neural networks, incligrad autograd) License: Creative Commons Attribution-NonCommercial-ShareAlike 4.0 (CC BY-NC-SA 4.0) RSS Feed: https://adarshdubey.com/rss.xml Sitemap: https://adarshdubey.com/sitemap.xml ## OWNER Name: Adarsh Dubey Bio: Full Stack AI Engineer with a passion for building in public and exploring the frontiers of LLMs. Recently completed GSoC '25 with Google DeepMind, currently exploring the tangents of AI from Tangent Lab. Core team member at the WeMakeDevs community. Expertise spans machine learning, natural language processing, and modern web development. Actively researching and experimenting with small language models, fine-tuning techniques, MCP servers, and efficient inference methods. Speaks at tech meetups, writes technical blog posts, and when not coding, plays badminton or enjoys misal pav. Notable open source projects: - Facet: An end-to-end Gemma fine-tuning platform - Incliarray: PyTorch-like N-dimensional float array library in C++ - Anon: Anonymous discussion platform for privacy-focused communities - Inclinet: Neural network library from scratch in Python - FastAPI MCP Inspect: Mount MCP servers onto FastAPI applications - Inclitoken: BPE tokenizer implementation from scratch - pplx-cli: CLI tool for the Perplexity AI API Philosophy: Strong advocate for open source, developer education, and sharing knowledge through technical blog posts and community contributions. Contact: - GitHub: https://github.com/inclinedadarsh - LinkedIn: https://linkedin.com/in/dubeyadarsh/ - Twitter/X: https://x.com/inclinedadarsh - Website: https://adarshdubey.com - Cal.com: https://cal.com/adarshdubey ## EXPERIENCE Current: - Tangent Lab: Experimental one-person AI lab (May 2026 — Present) Website: https://tangentlab.xyz Past: - Google DeepMind: GSoC '25 contributor (May 2025 — Sept 2025) Website: https://deepmind.google - WeMakeDevs: Web developer & core team member (2023 — 2025) Website: https://wemakedevs.org ## PORTFOLIO PROJECTS 1. Facet: Gemma Finetuner Description: End-to-end platform that lets users preprocess datasets, fine-tune Gemma models, run inference, and export them in different formats. Technologies: FastAPI, Python, Next.js, Hugging Face, GCP Links: https://github.com/gemma-facet 2. Incliarray: N-dimensional Array Library Description: PyTorch-like minimal N-dimensional float array library in C++ with row-major layout, stride-aware indexing, broadcasting, matmul, reductions, and lightweight reverse-mode autograd. Technologies: C++ Links: https://github.com/inclinedadarsh/incliarray 3. Anon: Social Media Forum Description: Anonymous social platform using optimistic UI with college verification and no personal data storage. Technologies: Next.js, FastAPI, Python, AWS, PostgreSQL Links: https://github.com/inclinedadarsh/anon | https://anon.adarshdubey.com 4. Inclinet: Neural Network Library Description: Neural network library from scratch in Python with PyTorch-like APIs, supporting multiple activation functions, losses, and optimizers. Technologies: Python Links: https://github.com/inclinedadarsh/inclinet ## EXPERIMENTAL / TOY PROJECTS 1. FastAPI MCP Inspect Description: Mount an MCP server onto your FastAPI application to let AI agents inspect your routes, endpoints, and schemas at runtime. Technologies: Python, FastAPI Links: https://github.com/inclinedadarsh/fastapi-mcp-inspect | https://pypi.org/project/fastapi-mcp-inspect/ 2. Inclitoken: BPE Tokenizer Description: Byte Pair Encoding tokenizer implementation from scratch in Python with custom training and encode/decode capabilities. Technologies: Python Links: https://github.com/inclinedadarsh/inclitoken 3. pplx-cli Description: CLI tool for the Perplexity AI API — web search, AI chat, streaming, citations, and support for multiple models (Sonar, Sonar Pro, Reasoning). Technologies: TypeScript Links: https://github.com/inclinedadarsh/pplx-cli 4. K-Means Clustering in C Description: Educational implementation of k-means clustering algorithm in C using a custom Matrix struct. Technologies: C++ Links: https://github.com/inclinedadarsh/k-means-c 5. Logistic Regression in C Description: Modular logistic regression with forward/backward pass approach, mimicking neural network style — all in C. Technologies: C++ Links: https://github.com/inclinedadarsh/logistic-regression-c 6. Neural Network in C Description: Basic neural network implementations including linear neuron (perceptron), non-linear single-layer network (OR gate), and multi-layer network (XOR gate). Technologies: C++ Links: https://github.com/inclinedadarsh/neural-network-in-c 7. Incligrad: Autograd Engine in C++ Description: PyTorch-like autograd library in C++ from scratch, including full backpropagation logic for gradient-based learning. Technologies: C++ Links: https://github.com/inclinedadarsh/incligrad ## BLOG POSTS (18 articles) Latest posts: 1. How I Built an MCP Server for FastAPI (2026-06-28) 2. Building A BPE Tokenizer from Scratch (2026-01-30) 3. Building a Newsletter App with Next.js and AutoSend (2025-11-23) 4. GSoC 2025 Project Report: Building Gemma Facet (2025-09-14) 5. The Small Language Model Revolution (2025-08-22) 6. GSoC Midterm: Project Overview and Updates (2025-07-24) 7. Learning Batch Processing the Hard Way (2025-06-25) 8. Meeting My Peers at Google DeepMind (2025-06-02) 9. Landing GSoC 2025 with Google DeepMind (2025-05-14) 10. Nights & Weekends Chronicles: Week 1 (2024-04-15) 11. College clubs and events — Are they even worth it? (2024-04-11) 12. First Steps Towards Learning Machine Learning (2025-01-28) 13. Embracing Syncloop's Simplicity (2023-07-30) 14. Seamless Code Formatting: ESLint, Prettier, Husky & Lint-staged (2023-07-01) 15. Comment Mind — Analyze YouTube Comments (2023-04-29) 16. Terminal Basics: A Comprehensive Guide for Beginners (2023-03-19) 17. Getting started with NumPy & NumPy Arrays (2023-01-24) 18. UX Design laws every developer should know (2022-11-10) Full list: https://adarshdubey.com/blogs RSS: https://adarshdubey.com/rss.xml ## TECHNOLOGIES - Python, PyTorch, Hugging Face - C++ - React.js, Next.js, TypeScript - FastAPI, PostgreSQL - AWS, GCP, Docker - Gemini, Claude - Git ## AI USAGE POLICY Allowed: Yes, with attribution Conditions: - Attribution to Adarsh Dubey and https://adarshdubey.com - Non-commercial use only - Share-alike for derivative works Not Allowed: Commercial use without explicit permission Contact for permissions: DM on Twitter/X (@inclinedadarsh) or GitHub ## TECHNICAL Stack: Next.js, React, TypeScript, Tailwind CSS CMS: Markdown files stored in src/blogs/ Update frequency: Regular (blog posts ~monthly, commits frequent)