Roam

2026-06-27
Original implementation credits go to Hugo Cisneros

Here lie bottom-up (not top-down) notes.

Flask

FastAPI

Hugging Face Agents Course

Unit 1. Introduction to Agents

I think as a rule of thumb, when using LLMs, you should just be grateful that it is able to understand LANGUAGE and nothing more.

You should not ask it the weather, nor should you ask it to play chess against you, because it does not know how to do those things – instead those activities (arithmetic, weather checking, chess-playing) should be delegated to TOOL CALLS!

Read more >

Agentic AI

I am trying to decode the agentic AI frameworks:

(Burtenshaw, Ben and Thomas, Joffrey and Simonini, Thomas and Paniego, Sergio, 2025)

LlamaIndex

LangGraph

LangChain

Crew AI

Pydantic AI

smolagents

Github: Ollama Voice Chess

Voice-controlled chess using Ollama.

ollama-voice-chess

https://github.com/abaj8494/ollama-voice-chess

Notes

Backend

the tech-stack used FastAPI on the back-end. usually I use Flask for everything, but the requirements of this project are asynchronous, real-time requests.

FeatureFastAPIFlask
SpeedVery fast (async, based on Starlette)Slower (sync by default)
Async supportNative async/awaitRequires extensions
Type hintsRequired, powers validationOptional
Auto docsBuilt-in Swagger UI at /docsManual setup
ValidationAutomatic via PydanticManual or extensions
WebSocketsBuilt-inRequires Flask-SocketIO

Frontend

the front-end was refactored from one huge index.html file into a Svelte front-end served with Vite.

Read more >