The problem LeoDevika solves
Problem
- Lack of support & Docs →
- painful experience building →
- give up onboarding & churn
- Tired of answering like ‘Broken Record’ :
- valueable hours wasted
- stop answering at all
Solution : AI Agent for Debugging & Troubleshooting
- Agent debugs first and weed out ‘duplicate’, ‘stupid’ questions.
- devs work on ‘actual’ problems - incentivized with bounty
- knowledge created is stored in, auto-updating documentation
Added Benefits of Easier Building
- Better Code Assistant for leo
- There’s not enough data & no model for Leo code assistant.
- When builders use LeoDevika, it would create new quality data to train leo with, leading into better code suggestions
- Auto-updating documentation
- the question tickets would show which part was painful or not explained enough
- when knowledge base is ripe, agent could attempt to add lacking explanation
- needs human copilot- it would be much faster and accurate then human alone
- RAG system needs knowledge-base summarization for optimization anyway, so its added synergy
Create positive feedback loop.
easier hacking → more devs onboarded → more product built → more data → better code assistant → even easier hacking
Challenges we ran into
We pivoted from original idea (ML Library transpiler) just a minute ago, due to the fact that we weren't good enough to crack leo syntax. I believe we could've done it if there were more onboarding support or docs for us - so we decided to create one ourselves.
For LeoDevika, it's quite standard - lack of LEO code data, GPU resource shortage, and general time limit.