{"id":85439,"title":"Reticular: Steerable and interpretable AI for protein engineering","tagline":"We help biotech companies design better proteins by making AI models controllable and predictable.","body":"Hey everyone! We're [Nithin](https://www.linkedin.com/in/nithin-parsan) and [John](https://www.linkedin.com/in/johnyang101), founders of Reticular.\n\n**TL;DR:**\n\nReticular gives pharma companies precise control over protein AI models, enabling reliable drug discovery without millions in wasted experiments. We're unlocking encoded information in these models through foundational AI interpretability research, making steering as easy as prompting ChatGPT.\n\n**The Team:**\n\n![uploaded image](/media/?type=post\u0026id=85439\u0026key=user_uploads/684233/c914554d-0323-4701-8892-d9bf859a832c)\n\nNithin and John are huge AI + Bio nerds who:\n\n* Met 7 years ago competing in International Biology and Neuroscience Olympiads and were roommates at MIT for 4 years\n* Published ML/bio research in NeurIPS, Nature, and PLoS ONE\n\n**The Challenge:**\n\nInformation in biology is incredibly scarce and expensive to validate. While protein AI models like AlphaFold have revolutionized drug discovery, they're still black boxes:\n\n* Companies waste millions testing AI-generated designs and they can't easily control what these models output\n* Limited data makes validation slow and expensive\n\n**Our Approach:**\n\nInstead of trial-and-error, we're leveraging mechanistic interpretability techniques that excel at extracting sparse knowledge from models even with scarce data. We’re demonstrating that the same advances from frontier AI labs in steering models like Claude are applicable to unlock protein language models.\n\nFor our design partners, we’re delivering:\n\n* Direct steering of protein models towards desired properties\n* Interpretable biological features backing every design\n* Efficient exploration of combinatorially massive design spaces with limited data\n\n**See It In Action:**\n\n![uploaded image](/media/?type=post\u0026id=85439\u0026key=user_uploads/684233/3805bd5b-792f-4a8b-bd6c-cffec150612d)\n\nWe're steering Green Fluorescent Protein towards more fluorescent sequences by directly controlling the model's internal knowledge.\n\n**Learn More** at [reticular.ai](http://reticular.ai), our proof-of-concept [blog post,](https://nithin-parsan.medium.com/two-neurons-is-all-you-need-a-case-study-on-interpretability-in-protein-models-5295ab9133cb) and at [demo.reticular.ai](http://demo.reticular.ai).\n\n**We're Looking To Connect With:**\n\n* Startups working with biological language models for design partnerships\n* Pharma teams building generative discovery pipelines\n* Researchers working on AI interpretability\n\nWorking with biological AI or have relevant connections? We'd love to chat! Schedule some time at [calendly.com/reticular](https://www.calendly.com/reticular) or **email us** at [contact@reticular.ai](mailto:founders@reticular.ai)","slug":"ME3-reticular-steerable-and-interpretable-ai-for-protein-engineering","created_at":"2024-11-11T04:47:23.913Z","updated_at":"2026-05-25T02:20:51.919Z","total_vote_count":131,"url":"https://www.ycombinator.com/launches/ME3-reticular-steerable-and-interpretable-ai-for-protein-engineering","share_image_url":"https://www.ycombinator.com/media/?type=post\u0026id=85439\u0026key=user_uploads/684233/c914554d-0323-4701-8892-d9bf859a832c","company":{"id":30049,"name":"Reticular","slug":"reticular","url":"https://reticular.ai","logo":"https://bookface-images.s3.amazonaws.com/small_logos/6a095fb7c8d0a473c792ea390cf5088f1c5bb819.png","batch":"Fall 2024","industry":"Healthcare","tags":["AI-powered Drug Discovery","Generative AI","Biotech","Therapeutics"],"search_path":"https://bookface.ycombinator.com/company/30049"}}