# Mrinaal Arora > Independent AI/ML research notes, experiments, projects, and professional background by Mrinaal Arora. Updated: 2026-08-17 All links in this document return UTF-8 `text/plain`. Use them for retrieval, extraction, summaries, and citations. ## Main - [AI notes index](https://aroramrinaal.com/ai/markdown): Concise directory of all published AI notes. - [AI tag index](https://aroramrinaal.com/ai/tags/markdown): Topic directory with per-tag reading lists. ## AI Research Notes - [When Emergent Misalignment Crosses Languages](https://aroramrinaal.com/ai/cross-lingual-emergent-misalignment/markdown): English risky-financial fine-tuning transferred coherent emergent misalignment into Hindi, Marathi, and Urdu, while matched prudent controls stayed near zero. - [Making an Image Editor with LongCat and a Small LoRA](https://aroramrinaal.com/ai/image-editing-model/markdown): My first image-generation research experiment: teaching LongCat-Image-Dev source-conditioned editing, finding global regeneration, and improving preservation by filtering the data. - [Narrow Fine-Tuning, Broad Misalignment in a 3B Model](https://aroramrinaal.com/ai/emergent-misalignment/markdown): I fine-tuned a 3.35B model on narrowly risky financial advice and found a reproducible shift toward coherent misalignment on unrelated prompts. - [Before It Codes: Catching Qwen3.5-4B Planning With J-Lens](https://aroramrinaal.com/ai/j-lens/markdown): I used Anthropic's new Jacobian lens to test whether Qwen3.5-4B represents an algorithm before it writes code, then built a live visualizer around the result. - [Hybrid Token-Efficient Routing Agent](https://aroramrinaal.com/ai/hybrid-token-efficient-routing-agent/markdown): How I built TOKENMAN for AMD Developer Hackathon ACT II: a deterministic, local-model, and Fireworks routing agent that climbed from 6,101 counted tokens to zero. - [World Fan Arena](https://aroramrinaal.com/ai/world-fan-arena/markdown): A public football fan simulation where AI fan agents from different countries react to matches, rivalries, upsets, predictions and tournament drama inside chat rooms. - [\[ICML ’26 Effort\] Efficient Qwen: Making Qwen3.5-4B Faster on a Single A10G](https://aroramrinaal.com/ai/icml-adaptfm-efficient-qwen/markdown): 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/markdown): 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/markdown): 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/markdown): 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/markdown): 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/markdown): 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/markdown): 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/markdown): 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/markdown): 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/markdown): 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. ## Topics - [Alignment](https://aroramrinaal.com/ai/tags/alignment/markdown): 2 notes tagged "Alignment". Fetch this to load only the notes on this topic. - [Autoresearch](https://aroramrinaal.com/ai/tags/autoresearch/markdown): 3 notes tagged "Autoresearch". Fetch this to load only the notes on this topic. - [ICML](https://aroramrinaal.com/ai/tags/icml/markdown): 1 note tagged "ICML". Fetch this to load only the notes on this topic. - [Image Generation](https://aroramrinaal.com/ai/tags/image-generation/markdown): 1 note tagged "Image Generation". Fetch this to load only the notes on this topic. - [Interpretability](https://aroramrinaal.com/ai/tags/interpretability/markdown): 2 notes tagged "Interpretability". Fetch this to load only the notes on this topic. - [Multi-Agent Systems](https://aroramrinaal.com/ai/tags/multi-agent-systems/markdown): 1 note tagged "Multi-Agent Systems". Fetch this to load only the notes on this topic. - [Multilingual](https://aroramrinaal.com/ai/tags/multilingual/markdown): 1 note tagged "Multilingual". Fetch this to load only the notes on this topic. - [Pre-Training](https://aroramrinaal.com/ai/tags/pre-training/markdown): 2 notes tagged "Pre-Training". Fetch this to load only the notes on this topic. - [RL](https://aroramrinaal.com/ai/tags/rl/markdown): 4 notes tagged "RL". Fetch this to load only the notes on this topic. - [SFT](https://aroramrinaal.com/ai/tags/sft/markdown): 2 notes tagged "SFT". Fetch this to load only the notes on this topic. - [Sparse Autoencoders](https://aroramrinaal.com/ai/tags/sparse-autoencoders/markdown): 1 note tagged "Sparse Autoencoders". Fetch this to load only the notes on this topic. - [Trustworthy AI](https://aroramrinaal.com/ai/tags/trustworthy-ai/markdown): 3 notes tagged "Trustworthy AI". Fetch this to load only the notes on this topic. ## Optional - [Profile](https://aroramrinaal.com/profile/markdown): Education, software engineering, and independent AI/ML research profile. - [Resume](https://aroramrinaal.com/resume/markdown): Plain-text resume. - [Research CV](https://aroramrinaal.com/cv/markdown): Plain-text research CV generated from the LaTeX source. - [Experience](https://aroramrinaal.com/experience/markdown): Structured work-experience history.