Github Link: https://github.com/rahulb99/meet-and-greet
Background
Came across this cool blog that included some end-to-end (almost production worthy) AI projects, with a focus on learning. So I decided to try out one of the projects called PhiloAgents, an agentic RAG application that allowed you to chat with some famous philosophers. It was fun learning this course, but since most of the code was already implemented, I wanted to get my hands a little dirty, so I made small changes to different parts of the codebase.
Thoughts
- Not a big fan of LangChain, especially after having used Agno and bit of LlamaIndex, I found it a bit cumbersome to use and the framework itself is so bloated.
gpt-oss-20bhallucinates like crazy.
- I was planning to add a TTS (text-to-speech) feature to the app via KittenTTS, a lightweight model, but its quality is only sufficient for basic use cases. Perhaps, I can add ElevenLabs when I get some free credits.
- The evaluation part with the synthetic dataset generation was interesting, something which I have also used in my own research projects on AI safety.