---
title: Autoresearch
description: AI notes by Mrinaal Arora tagged "Autoresearch"
updated: 2026-07-22
canonical: https://aroramrinaal.com/ai/tags/autoresearch
source: https://aroramrinaal.com/ai/tags/autoresearch.md
alternate: https://aroramrinaal.com/ai/tags/autoresearch/markdown
---

# Autoresearch

AI notes, small builds, and writeups by Mrinaal Arora tagged "Autoresearch".

- Page: https://aroramrinaal.com/ai/tags/autoresearch
- Markdown: https://aroramrinaal.com/ai/tags/autoresearch/markdown
- Markdown (.md): https://aroramrinaal.com/ai/tags/autoresearch.md
- All tags: https://aroramrinaal.com/ai/tags/markdown
- Full AI index: https://aroramrinaal.com/ai/markdown
- Posts with this tag: 3 of 15
- Co-occurring tags: ICML, Pre-Training

## Posts

| Post | Date | Tags | Word count | Source |
| --- | --- | --- | --- | --- |
| [Hybrid Token-Efficient Routing Agent](https://aroramrinaal.com/ai/hybrid-token-efficient-routing-agent) | 2026-07-22 | Autoresearch | 1,856 words | [Markdown](https://aroramrinaal.com/ai/hybrid-token-efficient-routing-agent/markdown) |
| [[ICML ’26 Effort] Efficient Qwen: Making Qwen3.5-4B Faster on a Single A10G](https://aroramrinaal.com/ai/icml-adaptfm-efficient-qwen) | 2026-06-30 | ICML, Autoresearch | 1,776 words | [Markdown](https://aroramrinaal.com/ai/icml-adaptfm-efficient-qwen/markdown) |
| [Parameter Golf: My OpenAI Model Craft Experiment](https://aroramrinaal.com/ai/parameter-golf-openai-model-craft) | 2026-05-04 | Pre-Training, Autoresearch | 2,034 words | [Markdown](https://aroramrinaal.com/ai/parameter-golf-openai-model-craft/markdown) |

## Hybrid Token-Efficient Routing Agent

- Slug: hybrid-token-efficient-routing-agent
- Date: 2026-07-22
- Tags: Autoresearch
- Word count: 1,856 words
- Page: https://aroramrinaal.com/ai/hybrid-token-efficient-routing-agent
- Markdown: https://aroramrinaal.com/ai/hybrid-token-efficient-routing-agent/markdown
- Markdown (.md): https://aroramrinaal.com/ai/hybrid-token-efficient-routing-agent.md
- GitHub: https://github.com/aroramrinaal/hybrid-token-efficient-routing-agent

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.
## [ICML ’26 Effort] Efficient Qwen: Making Qwen3.5-4B Faster on a Single A10G

- Slug: icml-adaptfm-efficient-qwen
- Date: 2026-06-30
- Tags: ICML, Autoresearch
- Word count: 1,776 words
- Page: https://aroramrinaal.com/ai/icml-adaptfm-efficient-qwen
- Markdown: https://aroramrinaal.com/ai/icml-adaptfm-efficient-qwen/markdown
- Markdown (.md): https://aroramrinaal.com/ai/icml-adaptfm-efficient-qwen.md
- X: https://x.com/arora_mrinaal/status/2067589267883470994

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.
## Parameter Golf: My OpenAI Model Craft Experiment

- Slug: parameter-golf-openai-model-craft
- Date: 2026-05-04
- Tags: Pre-Training, Autoresearch
- Word count: 2,034 words
- Page: https://aroramrinaal.com/ai/parameter-golf-openai-model-craft
- Markdown: https://aroramrinaal.com/ai/parameter-golf-openai-model-craft/markdown
- Markdown (.md): https://aroramrinaal.com/ai/parameter-golf-openai-model-craft.md
- X: https://x.com/arora_mrinaal/status/2046931839672340880
- GitHub: https://github.com/aroramrinaal/parameter-golf/tree/mri/lab-setup

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.