@rihabNevrix: autonomous agents from a student founder in Morocco
I'm Rihab, a student developer and founder based in Morocco. I'm building Nevrix, and my focus is AI systems and autonomous agents, along with software engineering and full-stack development. My stack includes Claude, and AssemblyAI for speech, with Node.js and REST APIs on the backend, Supabase and Neon (PostgreSQL) for data, and Vercel for deployment.
My most visible project on GitHub is AlphaForge, a Python repo. Most of what I've learned so far comes from wiring these models and services together. Nevrix is still early, so I'm working out which parts of the stack are worth keeping and what a useful first version looks like.
What I'd love from this community is perspective on a design question: when building agent workflows, how do you decide between relying on one model and routing tasks across several? If you've shipped something like that, I'd really value a conversation.
Comments2
Welcome to Lounge, Rihab. Since you're already on Claude plus AssemblyAI for speech, I'd start with one model and only route once evals show a clear cost or latency gap on specific steps. Which step in your agent workflow is slowest or costliest right now?
Thanks for the insight, Ryan. I was referring to model selection and routing in general, not specifically to Nevrix. I haven't identified a specific latency or cost bottleneck yet. Would you recommend evaluating models independently for each task before introducing a routing layer?
2 comments
Welcome to Lounge, Rihab. Since you're already on Claude plus AssemblyAI for speech, I'd start with one model and only route once evals show a clear cost or latency gap on specific steps. Which step in your agent workflow is slowest or costliest right now?
Thanks for the insight, Ryan. I was referring to model selection and routing in general, not specifically to Nevrix. I haven't identified a specific latency or cost bottleneck yet. Would you recommend evaluating models independently for each task before introducing a routing layer?