Open-Source AI Weekend Challenge: Build for a Friend

Shelf-life has been determined the same way for 50 years. Today we spoke alongside SGS about the future of shelf-life, at the THE SOCIETY OF FOOD HYGIENE AND TECHNOLOGY's AI Conference To reduce waste, keep food affordable and maintain high quality, the food industry needs richer data to make more precise shelf-life decisions. Currently shelf-life is determined using annual tests in simulated supply-chain conditions, then occasionally changed reactively (during summer heat waves). The reality is supply-chains are highly varied. An exciting new chapter is coming, where we surpass human instinct with AI and sensor infrastructure to underpin shelf-life decisions.

BlakBear is helping supply chains move from reactive to proactive, catching issues before they become waste, complaints, or lost margin.

We spend a lot of time talking about connected data. This weekend at the Meat Industry Expo, we're showing what that looks like in practice. In this short demo, Brandon Drung walks through Cesy, the new CSB AI Assistant. Cesy connects to your CSB data and helps answer operational questions in seconds, making information easier to access for the people who need it. Stop by Booth 130 if you're attending the show. We'd be happy to show you more or contact us today! #MPO2026 #MeatIndustryExpo #FoodProcessing #CSBSystem #AIAssistant

🧱Stage 8/9 — zooming all the way out. After spending the last post buried in one specific bug, I felt like the right time to step back and map the entire system in one diagram instead of just the piece that was broken. This is the full flow, start to end: photo comes in, gets preprocessed and routed to the right CNN (Indian or global), a confidence threshold decides whether to trust the prediction or flag it as non-food, nutrition gets pulled from SQLite (with a local-DB check before falling back to a dish not supported message), calorie targets get calculated against the user's profile and diet goal, history gets logged, the dashboard updates live, and an AI insight tip gets generated based on remaining calories. Off to the side, Quick Ask runs its own path — checks for internet, calls Gemini if available, falls back to local logic if not. Laying it all out like this actually caught a few small inconsistencies in how Id been describing the system to people asking questions in earlier posts — good reminder that a diagram forces a kind of honesty a verbal explanation doesn't. Last post in the series next — pushing this to GitHub, getting teammates running it locally, and closing out with where the project actually stands now. #ai_powered_food_calorie_estimator

We're getting to grips with our latest, self funded AI qual study on GLP-1 users - 200 respondents, 5 leading edge markets, 450000 words of testimony fresh from the lips of real consumers. The results are fascinating. GLP-1 has rewired how consumers think about food. Most consumers perceive a tension between 'good' behavior - eating well and exercising - vs 'bad' behavior - sitting on the couch eating Doritos. In effect eating well is a game at which many consumers, particularly in the US, do not succeed. GLP-1 seems to be a 'chemical breach' in that framing. Since 2023 consumers have started to talk about 'food noise' being the problem, and GLP-1 being the solution. Suddenly food is the bad guy and GLP-1 is a magic wand which takes away it's powers. Unsurprisingly, this removes a ton of guilt and allows users to re-evaluate many aspects of their life. If you want to find out more drop our awesome new qual team a line on Qual.SAI@smb.nielseniq.com and we can tell you all about it.

Fascinating work from Joe and the team. The impact of GLP-1s extends far beyond food and beverage, creating ripple effects across many Non-Consumables categories including retail, health & wellness, home, technology, automotive, and beyond. Understanding how consumers are reframing habits, priorities, and decision-making requires getting beyond the data to uncover the "why" behind behavior. This is a great example of how our qualitative team can help companies anticipate change, identify emerging opportunities, and make smarter business decisions in rapidly evolving markets. A special shoutout to Joey Doney, who leads our North America Qualitative practice and is helping clients unlock these types of deeper human insights every day. Whether you're looking to understand emerging consumer behaviors, pressure-test innovation, explore brand perceptions, or uncover new growth opportunities, don't hesitate to reach out to Joey or me.

AgriCuke - AI cucumber-farming assistant we built with Flutter + TensorFlow Lite. Farmers photograph a leaf, get an on-device disease diagnosis in under 2 seconds, and sync field readings when they're back on signal. Offline-first was the whole design brief. https://lnkd.in/e_JcvBFa

The episodes where I walk through how I actually use AI in everyday life are consistently our most popular, so they’re now a monthly feature. This month’s 13 range from the practical (cleaning Sunbrella fabric, troubleshooting a Nest thermostat before calling an HVAC company, pulling square footage out of an apartment PDF) to the personal (helping my sister get her passwords in order, and a reading partner for a long, complex novel). I use Claude, but nearly all of it works with any AI assistant. Also in this episode: The Inner Game of Tennis becomes an AI coaching app; Steven Rattner New York Times essay, “Six Charts That Show Just How Much We Need A.I.”; and, for AI for Good, Clairity, Inc. Breast, an FDA-approved AI test that predicts five-year breast cancer risk from a single mammogram. Listen or view wherever you get your podcasts. #AIforSeniors #ArtificialIntelligence #ClaudeAI #aiGED #AItips #SeniorLiving #EverydayAI https://lnkd.in/gUThCb5g

I tested Jev, the new decision / classification model from TypeSafe AI, against gnarly response time issues I've been trying to crack for GlutenOrNot (which uses Opus 4.8) and got some pretty awesome results. - Jev's median response time was 17x faster - Jev's cost was ~240x cheaper across 1,000 ingredient labels - Did not miss a single label that visibly contained gluten We're going to see products bring in these super fast/cheap classifier models immediately. The cost savings / speed gains are wild. Full writeup (with example experiments repo you can test with) here -> https://lnkd.in/gyaMuqqG

Our very own Andrew Larty presented “From Signal to Surveillance: How AI is Automating Pharmacovigilance at Scale” to attendees at PCC West this week. Behind every safety event is a patient who trusted someone to listen. Authenticx has built a Pharmacovigilance AI Solution that prioritizes patient safety with precision and speed. If you want to learn more about what Andrew talked about today in San Diego, we’d love to chat. Informa Connect Life Sciences https://lnkd.in/gmVYur8D