Night-shift log: Jev works best for simple classification or probability handoff
After weeks of running Jev in his pipelines, the developer concludes it is only good at simple classifications or as a probability handoff that narrows the scope for a real LLM.
3hr+ timelapse, doing Night shifts working for US clients: > Reviewed some PRs for a chatbot, tons of bugs and tons of issues. > Added the social sensing pipeline to fly[.]io running weekly scheduled jobs. Monitoring these jobs with Langfuse + Healthchecks + Bugsink. > Did some Analysis on how Jev was working from the past few weeks in my pipelines - Ended up with the conclusion that it is only good at simple classifications or as a probability handoff to actual LLM to reduce the scope of work for LLM. > Added some web search normalization, shifted from using Parallel to Exa as the tool call layer for web search for one of the agents. > Made an Admin Dashboard in a client's project with AshAdmin and PhoneixLiveView. > Started working again on my brain memory project for sub agents after a long time. > Also I setup Deskflow this week as the software-KVM for my mac and windows so that I can slide in my cursor across the 3 screens now. Orca + windows SSH works great too. > Refined the Marketing and Job search pipeline for one of the clients. Added some more slack commands for the their team. > Did some research to understand the LiteLLM Supply Chain Malware attack cuz we were using litellm in one of our pipelines. That's it - Peace out !!