Roadmappy swaps LLM classification for Jev in its core workflow
A developer's weekly retro says Roadmappy's raw input to signal to feedback to plan workflow now uses Gemini's latest embedding model plus Jev for classification instead of the previous LLM approach, taking longer but with great results.
Good morning folks! It's a lovely sunny day today ☀️ Can even see the alps in the background. Another week, another set of goals. What are your weekly plans? Wishing you you all a great week! My plan for this week: • Succeed in 9-6, but only till Tuesday • Go on holiday on Wednesday (easy to achieve 😆) • Get a secret project live 🤫 • Re-work Roadmappy monetisation and onboarding journey • Reach out to more initial testers of LimeGhost • Improve the AI agent experience in LimeGhost • Start video production for YouTube • Run 30km What I said I would do last week (retro): • Succeed in 9-6 ✅️ • Give a great speech about AI transformation and network with many people ✅️ • Overhaul the core workflow in Roadmappy: raw input > signal > feedback > plan. Need to update to Gemini's latest embedding model and switch to Jev for classification (instead of my still used LLM approach) ✅️ (this took a long time, but results are great) • Overhaul Roadmappy website ✅️ • Get approved by Google for Youtube integration from LimeGhost ‼️ (did not start) • Integrate Tiktok into LimeGhost ‼️ (did not start, de-prioritised) • Run 30km ‼️ (ran 20km)