Entelect

Last Saturday we hosted our second GirlCode Group x Entelect workshop of the year👩💻 A big thank you to our speakers: - Vanessa Mohr for an insightful and practical Python session - Jessica Ugboaja for simplifying Version Control - Ayanda More for sharing the expectations vs reality of breaking into and working in the tech industry We were also fortunate to have Zandile Mkwanazi , CEO of GirlCode, join us!🥳 It’s always rewarding to see so many engaged and motivated ladies come together to learn, build skills, and support one another🫶 Also, a special thanks to Renette Ros , Jenique P. , Divya Soomaroo , and Gertruida Maritz for assisting on the day with questions and debugging support🐛 If you’re interested in joining a future workshop, you can sign up at https://lnkd.in/dDJQcQ96 or reach out to GirlCode Group directly.

Entelect UK's Tech accelerator - an initiative for Entelectuals to work on their own projects, explore cutting-edge technologies, tackle challenges, and develop future products beyond their day-to-day responsibilities. For the second hackathon we had Dean Cowley , Anthony Farquharson and myself working through two projects coupled with AI discussions, of course. To reiterate; The first project - Kamber - Dean Cowley Kamber is a collaborative quality orchestration platform that bridges the gap between business requirements and source control by turning "Three Amigos" workshops into automated technical footprints. It's the structural angle needed to eliminate context-switching and keep fast-moving mobile sprints on track The second project - Clarity - Me A tiny transformer-based language model built from scratch in pure C++ that translates cryptic GCC/Clang compiler errors into clear, human-readable explanations. The insights from this project will contribute to TinyModels, an initiative I'm working on to create multiple specialised tiny models that can work together. Rather than building general-purpose models, each one is designed to do one thing, and do it exceptionally well. We also had a fascinating discussion on PolarQuant, a clever new quantisation technique that represents high-dimensional vectors (especially KV cache embeddings) in polar coordinates instead of traditional Cartesian ones. By separating magnitude from direction and applying smart per-component quantisation, it delivers massive memory savings with virtually no accuracy loss. This kind of innovation is exactly what makes truly lightweight, specialised models practical to run on everyday hardware. In my opinion. the future of models.