Biographica Selected for Norrsken 100 List for Deep Tech in Food Systems

I am pleased to share that our research paper has been published! Title: A Knowledge Distillation Framework for Lightweight Bean Leaf Disease Classification for Resource-Constrained Deployment Journal: Dutse Journal of Pure and Applied Sciences (DUJOPAS), Vol.12 No.3c, September 2026 In this work, we proposed *DistilNet*, a lightweight architecture for bean leaf disease diagnosis on low-end mobile devices. Using a Teacher-Student Knowledge Distillation framework (ResNet50 -> DistilNet) with Bayesian hyperparameter optimization. This demonstrates that we can build efficient, offline diagnostic tools for smallholder farmers without compromising accuracy. #KnowledgeDistillation #DeepLearning #PrecisionAgriculture #EdgeComputing #BayeroUniversityKano #AI #Research

Would you trust AI to tell you whether a wild mushroom is safe to eat? 🍄 📜In our recently published study, we tested AI-based mushroom identification tools using real-world photographs. Even the best-performing tool failed in almost 15% of cases, and none consistently provided a single correct identification. ❓ Beyond mushroom foraging, the study raises a broader question: how much should we trust AI when incorrect answers can have real-world consequences? 🤝 It was a pleasure contributing to this interdisciplinary research and exploring a topic where technology, biology and public safety intersect. 📎 Kuznetsov, N.V., Xia, Y., Kuznetsov, S., King, F.C. and Ljubisavljevic, M. (2026) ‘AI-mediated risks and real-life challenges in mushroom foraging’, npj Science of Food, 10, 121. https://lnkd.in/dtw9AAKd #ArtificialIntelligence #Research #ResponsibleAI #PublicHealth #InterdisciplinaryResearch

“Programmable Biology: Reading and Rewriting the Human Body” -- this is the panel at GoHub Ventures’ Vertex Summit next week that will feature Andrii Buvailo, Ph.D. in company of industry insiders like Alberto Conti (Alceus Bio), Jason Mills 🕵️ (Talvo), and Marisol Quintero (Columbus Venture Partners). For most of history, biology was something we observed. AI is turning it into something we can design. The panelists will dicsuss the meaning of this shift in practice, what is real today and what is still science fiction, and how programmable biology moves us toward prevention rather than reaction. The Vertex Summit is hosted by GoHub Ventures and is supported by Microsoft. More information here: https://lnkd.in/eUuMEs3r

Biology is approaching an inflection point where we can demonstrate a controllability of design that I could have only dreamed of during my synthetic biology days in academia. We owe this shift to the huge foundation models like EDEN and Evo2, trained on the increasingly large tree of life data, and will be furthered by the continued rapid scaling of these models. It was great to have our work mentioned in The Economist this week alongside the great work from our peers at The Arc, Biohub, Google DeepMind and Genyro Inc https://lnkd.in/eZ96CceV

DNA is the source code of every living system. At Living Models, we’re building its reader. 🧬 That was my message at the opening keynote of ai-PULSE 2026 at STATION F this week. I shared the stage with Zelda, whose team at Bioptimus is building M-Optimus, a model connecting what happens inside human tissue to how disease evolves. Aude Durand moderated. We are a frontier lab. Our job is to develop the models and methods needed to tackle the biggest challenges in biology. What they can unlock is already visible: in our recent work with Google DeepMind, pairing Gemma with our BOTANIC models helped biologists find the genes behind a trait. And with biological data becoming AI-ready, and the talent right here in Europe, there is a lot to be optimistic about. Thank you to Aude & Scaleway for the invitation, and to STATION F (🫶🏼 Roxanne) for being instrumental in Living Models’ journey. And to our biologists and AI researchers working side by side: this stage was yours too.

🔴 [LIVE] The momentum continues at ai-PULSE! During today's ai-PULSE session on AI for Science, Zelda Mariet (Co-founder and VP of Research at Bioptimus) and Léonard Strouk (Co-founder and CTO at Living Models) are unpacking what it takes to train AI on living systems rather than digital text. 🎤 Key themes discussed on stage: • Different Data, Different Rules: Biological models require DNA, pathology images and clinical outcomes, making data assembly (like Bioptimus's STELA project) the primary bottleneck. • Real-World Impact: Living Models showcased BOTANIC-1, demonstrating how plant biology models generalize across biological domains to build climate-resilient crops. • The Interpretability Frontier: In science, a "black box" prediction isn't enough; researchers need to understand why a model reached a conclusion to validate scientific discoveries. A powerful demonstration of how European innovation is pushing the boundaries of biological foundation models. 👏 👉 Live here: https://lnkd.in/er5Qd-XR #aiPULSE #European #Open #Pragmatic #BuiltForWhatsNext

Most AI systems ask you to trust them. I spent the last two years building ones you can check instead. Today we are putting our whole stack online, free to explore. the researches hub https://lnkd.in/gKm23UFr DSOS Science — the complete method behind Squirrel OS self-healing infrastructure, taught as an 18-chapter textbook plus study companion. Every claim is backed by a public DOI. The transparency ledger https://lnkd.in/g6uaJqF9 An append-only ledger of every experiment, cost, and decision. Live Zenodo feed, key findings pinned, nothing hidden. When we say self-healing, we show the healing logs and the timestamps The proof you can touch https://lnkd.in/gXjhvT4Q DeterminCode — code you can prove, not code you have to trust. Everyone else asks an AI for code and hopes its right. We make the AI show its work: every function becomes an explicit chain of operators you can check like an equation. Same code in, same chain out, every time, with a checksum to prove it. It runs the Squirrel OS self-healing layer itself. The receipts https://lnkd.in/gGBP56qR Five published test batteries — social engineering resistance, autophagy, dead-hub recovery, self-perception, and live-write validity — all with DOIs, raw evidence, and frozen specs.T he thesis is simple: an AI system that modifies itself needs a deterministic boundary around the parts that decide. We publish the boundary, the evidence, and the code. Come break something. Tell us what you find. #DeterministicSystems #SelfHealingAI #AIOps #AIAgents #OpenScience

🧬 Decoding Biology: What can AI learn from human tissues and plant genomes? We are very excited to reveal 2 new amazing speakers for ai-PULSE! 🔥 Biology presents AI with radically different kinds of information. Zelda Mariet (Co-founder & VP of Research at Bioptimus) and Léonard Strouk (Co-Founder and CTO at Living Models) will bring two perspectives to the discussion: connecting tissue images with molecular data and learning from plant DNA. Together, they’ll explore what it takes to build foundation models for living systems, where their approaches converge, and where biology demands different methods. Curious about the future of AI in biology and hard science? Don't miss this discussion! 👉 Check the full agenda and register here: https://lnkd.in/ezwqqrF6 📅 October 13, 2026 📍 STATION F, Paris #aiPULSE #European #Open #Pragmatic #BuiltForWhatsNext

𝗜𝗻𝘁𝗿𝗼𝗱𝘂𝗰𝗶𝗻𝗴 𝗶𝘁𝗲𝗿𝗔𝗜 𝗽𝗲𝗮𝗿𝗹𝘀 At iterAI we work every day with data from R&D and production: trials, batches, sensors, lab results. Along the way we see the same small things make a big difference, again and again. So we are starting iterAI pearls: short, practical tips you can use right away. Statistics and experimental design, building reliable datasets, how AI tools work and how to use them safely, and a few everyday tricks for the tools you already use. Written for anyone working with data, from the lab to the boardroom, in aquaculture, insects, microalgae, fermentation and beyond. First pearl coming up: statistical power, or why "no significant difference" does not always mean "no difference". #iterAIpearls #DataScience #RnD

I am pleased to share my research project, “FedAgri-XAI: A Federated Attention-Based Network with Explainable Diagnosis for Privacy-Preserving Plant Disease Recognition Under Non-IID Farm Data. ”The research combines Federated Learning, MobileNetV2, CBAM attention, and Grad-CAM to enable privacy-preserving plant disease recognition with explainable predictions. It compares FedAvg and FedProx under different data heterogeneity conditions and evaluates explanation faithfulness using deletion AUC, insertion AUC, and IoU. The experiments achieved approximately 99.6% accuracy under IID conditions, while FedProx outperformed FedAvg under challenging cross-crop non-IID conditions. This work explores the potential of trustworthy and privacy-preserving AI for precision agriculture.#ArtificialIntelligence #FederatedLearning #ExplainableAI #DeepLearning #PrecisionAgriculture