# Mrinaal Arora

This site exposes plain markdown routes for AI agents and programmatic readers. Prefer these markdown URLs over rendered HTML when available.

## Main Pages

- [AI notes index](https://aroramrinaal.com/ai.md): Index of AI notes, small builds, and research writeups, with links to each post markdown source.
- [Resume](https://aroramrinaal.com/resume.md): Markdown resume for Mrinaal Arora.
- [Experience](https://aroramrinaal.com/experience.md): Structured markdown version of the homepage experience cards.

## Human Pages

- [Home](https://aroramrinaal.com)
- [AI notes](https://aroramrinaal.com/ai)
- [AI projects](https://aroramrinaal.com/ai/projects)
- [Resume page](https://aroramrinaal.com/resume)

## AI Research Notes

- [[ICML ’26 Effort] Efficient Qwen: Making Qwen3.5-4B Faster on a Single A10G](https://aroramrinaal.com/ai/icml-adaptfm-efficient-qwen.md): My ICML 2026 AdaptFM effort to optimize Qwen3.5-4B for low-latency inference on a single NVIDIA A10G through 100+ runtime, compiler, and quantization experiments.
- [Looking Inside Qwen3-4B With Sparse Autoencoders](https://aroramrinaal.com/ai/looking-inside-qwen3-4b-with-sparse-autoencoders.md): I trained a sparse autoencoder on Qwen3-4B-Base, labeled its learned features, and tested one by steering the model toward cooking instructions.
- [Pre-Training My First Base Language Model From Scratch](https://aroramrinaal.com/ai/first-pre-train-llm.md): Took a model from random initialization through 2B tokens of FineWeb-Edu on a single H100 and watched it learn next-token prediction from nothing.
- [Parameter Golf: My OpenAI Model Craft Experiment](https://aroramrinaal.com/ai/parameter-golf-openai-model-craft.md): I didn't come close to the Parameter Golf leaderboard, but I still had a lot of fun running scattered H100 experiments on Modal, hunting tiny BPB improvements while watching ideas collapse against artifact size limits, and figuring out the hard way why squeezing a capable model into 16 MB is trickier than it sounds.
- [CrisisOps: My Final Round OpenEnv Hackathon Project](https://aroramrinaal.com/ai/crisisops-openenv-hackathon-finale.md): How I built CrisisOps for the OpenEnv hackathon finale, then trained a small model with GRPO using TRL, Unsloth, and Modal.
- [My First Successful RL Training Run With OpenEnv And Wordle](https://aroramrinaal.com/ai/rl-training-openenv-environment.md): A full fine-tune of Qwen3-1.7B on Wordle using OpenEnv, with a reward curve that actually went up and shaped rewards that taught real gameplay behavior.
- [My Second RL Experiment On Prime Intellect Lab](https://aroramrinaal.com/ai/second-rl-experiment-on-prime-intellect-lab.md): My second RL experiment while studying GRPO: moving to Prime Intellect Lab, testing smaller and larger setups, and finally getting a reward curve that went up.
- [First RL Experimental Project: RLVR Using GRPO With TRL On Modal](https://aroramrinaal.com/ai/first-rl-experimental-project.md): My first ever RL experiment: RLVR on GSM8K using Hugging Face TRL, Qwen2.5 1.5B Instruct, and an NVIDIA H100 on Modal, with notes on where the reward function broke and what I want to fix in v2.
- [My Current Approach in Learning and Experimenting with LLMs and Deep Learning](https://aroramrinaal.com/ai/my-current-approach-in-learning-and-experimenting-with-llms-and-deep-learning.md): How I study LLMs by going deep on specific topics instead of starting from math.
- [Teaching a 3B Base Model to Emit Reasoning Traces](https://aroramrinaal.com/ai/nanbeige-cold-start-sft.md): A first small-scale AI research experiment: cold-start SFT on Nanbeige4-3B-Base using 2,160 distilled reasoning triplets, trained with LoRA on an H100, with notes on what worked and what I want to try next.

## Source Of Truth

Use the `.md` routes for extraction, summaries, retrieval, and citations. Use the paired HTML routes for visual presentation.