Pranav

Just spent a week in SF with my co-founders in the new @slamco_ space. I’m feeling a confidence and focus I haven’t felt in a year. Building a business with passionate, generous and incredibly talented people has that effect! We are so back 📈

@juliapintar playbook

@jacobrodri_ ⚡️

@rom1trs Editor

Every-time I’m with my home friends they remind me of this 💀 I love the tech scene but it can feel overly opinionated and noisy sometimes Important to take diverse perspectives on things

AI memory was my first real engineering challenge for Entries Its a journalling app that 'remembers' everything you write It helps you reflect by asking questions and surfacing patterns in your thoughts, ideas and emotions. Making the journal feel truly personalised had me immediately wary of using things like rag or storing occasional important 'memories'. Such a system would have low resolution and in the case of a retrieval algorithm the response quality could be bottlenecked by the ability for the algorithm to find the correct things in the user's journal entries. So similarly to what @rauchg is describing, I've built a system that utilises the large context window of modern models as much as possible and compresses journal entries as they get older. IN addition to storing key memories about users. So based on the user's input, every response as as personalised as possible. Still madly tinkering with this system and building upon it, but clearly something's working right, I've scaled it to over 200k journallers and growing just this year! Search myentries[dot]ai , always looking for feedback :)