{"id":88188,"title":"Augento 🤖 - DeepSeek reinforcement fine-tuning as-a-service","tagline":"Improving your AI agents with reinforcement learning.","body":"Hey all 👋, we’re Linus, Hannes, Lukas, and Josef—co-founders of [**Augento**](https://www.augento.ai/)! 🕊️\n\n**TL;DR**\n\nWe align your agents with reinforcement finetuning. You give us your agent, tell us where it fails and we’ll improve it. 🚀\n\nWe are actively looking for **design partnerships**. If you are interested, please shoot us a message at [**founders@augento.ai**](mailto:founders@augento.ai) 😄\n\n**The Problem ❌**\n\n🗣️ AI Agents struggle in real-world environments. Even state-of-the-art reasoning models score below 50% accuracy on non-trivial benchmarks.\n\nThe solution for many is still prompt engineering \u0026 expanding the models' guardrails. However, anyone who’s fought with large prompts knows how draining it can be. You’re never quite sure if the model is really following your instructions or if your tweaks make any difference.\n\n**Our Solution ✅**\n\nWe tackle this by replacing prompt engineering with fine-tuning + RL on **your** feedback. We integrate with a two lines of code change in your existing system.\n\n[**https://www.youtube.com/watch?v=Sulzf7VZr3k**](https://www.youtube.com/watch?v=Sulzf7VZr3k)\n\n**The Workflow 💡**\n\n1\\. Swap out your LLM connector URL with ours.\n\n2\\. We intercept every prompt and output, displaying them in our UI.\n\n3\\. Where necessary, you give high-level feedback, like your preferred tone or how a tool should actually be used.\n\n4\\. We continuously post-train the model to match up to your feedback.\n\n5\\. Once you deem it good enough and want to switch over to the model, you can do that with a click of a button, no changes to your code required.\n\n**Our Ask**\n\nWe’d love your input and are looking for early users to test-drive Augento. Shoot us a message at [**founders@augento.ai**](mailto:founders@augento.ai)\n\n**The Team**\n\n![uploaded image](/media/?type=post\u0026id=88188\u0026key=user_uploads/1922635/a0ae804d-98ad-434e-a7ef-cddacb16ed11)\n\nLukas previously studied Data Science @ ETH Zurich and developed deep learning optimizers, improving SGD’s generalization performance across CV. During his studies, he worked as a software engineer.\n\nLinus previously studied CS @ ETH in Zurich and did research in complexity theory. During his studies, he worked as an ML Engineer \u0026 as a Quantitative Developer in High-Frequency Trading.\n\nHannes previously studied CS @ ETH Zurich and worked on decentralized and distributed systems. During his studies, he worked as a paid contributor for a big open source project and as the technical lead for his previous startup.\n\nJosef previously studied CS @ ETH Zurich, and worked on computer systems and networks, as well as ML. During his studies, he worked as a full-stack software engineer and embedded systems developer.","slug":"MwO-augento-deepseek-reinforcement-fine-tuning-as-a-service","created_at":"2025-03-02T02:49:45.257Z","updated_at":"2026-05-25T01:43:46.166Z","total_vote_count":18,"url":"https://www.ycombinator.com/launches/MwO-augento-deepseek-reinforcement-fine-tuning-as-a-service","share_image_url":"https://www.ycombinator.com/media/?type=post\u0026id=88188\u0026key=user_uploads/1922635/a0ae804d-98ad-434e-a7ef-cddacb16ed11","company":{"id":30264,"name":"Stillwind","slug":"stillwind","url":"https://stillwind.ai","logo":"https://bookface-images.s3.amazonaws.com/small_logos/c1b36aa3102c0461123d8bc84ffc46b58b33f3a9.png","batch":"Winter 2025","industry":"B2B","tags":["AIOps","Artificial Intelligence","Electronics"],"search_path":"https://bookface.ycombinator.com/company/30264"}}