{"id":108337,"title":"hiloop: we run thousands of experiments to improve your models","tagline":"Send us a hard task, your model or agent, and an eval. We run the research campaign and return a verified improvement.","body":"We're Karan and Thomas, and we're building hiloop: **infrastructure for automated research**.\n\n**TL;DR: we gave two stock coding agents 50 B200s and our infra. They ran 4,188 experiments in two days and beat the published state of the art on [Karpathy's autoresearch benchmark](https://github.com/karpathy/autoresearch). The same machinery found a [NanoGPT speedrun](https://app.primeintellect.ai/speedrun/nanogpt) candidate that beats the best known result. [See](https://hiloop.ai/blog/search-is-enough/) full writeup.**\n\nNow we want to point it at your hardest problems.\n\n![uploaded image](/media/?type=post\u0026id=108337\u0026key=user_uploads/1714115/63396f7f-3e21-41ea-83ea-54cbaaa9f63f)\n\n**The results**\n\nOn Karpathy’s autoresearch benchmark, our final recipe reached **0.9016 val_bpb**. The previous best published result was [Recursive’s 0.9109](https://www.linkedin.com/pulse/first-steps-toward-automated-ai-research-recursive-si-kybtc) on B200 hardware. For NanoGPT, our implementation had **12.54% lower runtime** than upstream.\n\nThere was no bespoke research agent or elaborate scaffold. We used Claude and GPT-5.5 out of the box. What changed was the system around them.\n\n**What hiloop does**\n\nGive us a hard task, your current agent or model, and an evaluation criteria. Hiloop runs an autoresearch campaign across training, data, prompts, tools, harnesses, and systems, then returns the best verified improvement. It runs in your cloud or hosted by us.\n\nWe’re starting with agent and model training: SFT, post-training, continual learning, and optimization.\n\n[https://drive.google.com/file/d/1Rv8HttSHZ-yvSJk6hsWQsqCXWTHJQogK](https://drive.google.com/file/d/1Rv8HttSHZ-yvSJk6hsWQsqCXWTHJQogK/view?usp=sharing)\n\n**Who we are**\n\nWe met at Reducto, where we built its ML, post-training, and platform systems. Karan previously led ML infrastructure at DynamoAI . Thomas was a founding engineer at Crosswise and an MLE at Discord and SoFi.\n\n**Two asks**\n\n1. **Want your models or agents better, faster, or cheaper?** If you have a hard objective in post-training, continual learning, inference optimization, or classic ML, send us the task, your baseline, and the eval. We’ll personally run a campaign with you through verified improvement.\n2. **Building autoresearch or RSI loops?** Reach out to give our system a try! We’d love to compare notes.\n\nEmail [**founders@hiloop.ai**](mailto:founders@hiloop.ai) to get in touch with us!","slug":"SBN-hiloop-we-run-thousands-of-experiments-to-improve-your-models","created_at":"2026-07-29T19:50:09.250Z","updated_at":"2026-09-19T06:51:46.144Z","total_vote_count":8,"url":"https://www.ycombinator.com/launches/SBN-hiloop-we-run-thousands-of-experiments-to-improve-your-models","share_image_url":"//bookface-static.ycombinator.com/assets/ycdc/yc-og-image-c440a0ad1dacfb86eeeb343717479cc54d256614449b4ef719977a0a451f8bc8.png","company":{"id":33348,"name":"hiloop","slug":"hiloop","url":"https://hiloop.ai","logo":"https://bookface-images.s3.amazonaws.com/small_logos/3757083c195e320f3b98de96aa489419197b9463.png","batch":"Summer 2026","industry":"B2B","tags":["Artificial Intelligence","Developer Tools","Machine Learning","Reinforcement Learning","Infrastructure"],"search_path":"https://bookface.ycombinator.com/company/33348"}}