People often underestimate startups that aren’t based in major tech hubs. Personally, I’ve seen how a lower cost of living has allowed startups like Stylo to bootstrap and extend their runway in ways that wouldn’t be possible in a big city.
#Midwest #Startup #BuildInPublic
If you figure it out - please let me know 😆
No better way to make sure that the new "Share" functionality on Scorecard works than doing it live.
Check out our Stylo Scorecard! 🦋 scorecard.askstylo.com/view/fd3e192...
#buildinpublic
A wins a win 😆
People should be making their websites / knowledgebases accessible for the visually impaired. But if it takes showing people that its good for AI too, at least it gets us a more accessible web for everyone.
www.linkedin.com/pulse/how-al...
#buildinpublic #accessibility #ai
Shadowing customers while they use our products has never been a waste of time. Every session reveals things I’d have never spotted otherwise.
Train station in the snow
Hey #buildinpublic friends - how do you do it? How do you make the time to post consistently? I feel like I spend all of my time just building and talking to customers. Then every time I think about posting - it doesn’t feel like a priority.
Our AI User Group is looking for more interviewees for our 2025 “State of AI” Report— so far we've completed 18 interviews. If you work in the customer support world and want your voice heard, comment below or connect with me!
#zendesk #salesforce #freshdesk
For the most part! Though we're finding that for memory its far more accurate to be represented using a knowledge graph vs a vector db.
I'm not sure if this one will be recorded, its going to be at a conference. I'll most likely do a version of it for my AI User Group which I will record :)
Thanks! This one will be on memory management of LLMs. Specifically methods to not only capture memory but the different facets of what should be captured and stored for a better experience using AI
Revamping our onboarding guides, fixed a gap in my testing suite, and started outlining a talk I’m giving in March.
Founders, how much do you prioritize customer docs and onboarding guides?
Starting to improve ours and it feels like a balancing act—keeping things well-documented while building so fast it’s easy to fall behind. How do you handle this?
#startups #buildinpublic
Been playing around with DeepSeek V3 and its wildly good. Will be interesting to dig deeper and see if the accuracy / quality is there to justify standing up a new rack to use it on a production level. Any one else finding success with it?
#AI #GenerativeAI #NLP #LLM #OpenSource
This kind of thing can make all the difference in helping keep a users attention!
Just finished up the V1 designs for Stylo Scorecard! It’s a monthly scorecard where people can sign up to see how well their Knowledge Base is set up to be used as a context source for AI.
Stylo AI 2024 recap
So proud of all we accomplished in 2024 at askstylo.com. My resolution in 2025 is to try to share more of our wins and struggles.
#buildinpublic
#ai
I think what I need to build is a tool to essentially identify these issues up front and score the KB on qualitative metrics. Could be important for RAG quality as well
- Overfitting (not a problem in my use case, but could be for general model builders)
- Interlinked content ( also an issue for RAG)
Main struggles with using a KB as a fine-tuning source:
- Limited dataset size.
- Content in KBs often needs validation to confirm it’s not stale
- Inconsistent formatting
- Images and Videos galore
- Lots of redundant content
- Domain specific terminology
#buildinginpublic
Prepping a dataset for a talk next week on fine-tuning an LLM with a knowledgebase. It’s awesome how layering RAG with a fine-tuned model boosts retrieval accuracy and response quality.
