jamesob
Featured
local-llm
A comprehensive guide for building and configuring a high-end local machine to run state-of-the-art LLMs, with detailed hardware choices, BIOS tuning, and Docker-based model serving.
AI-Debater
The Floor is an MVP AI debate opponent that runs without paid APIs: you pick a category, spin for a topic, choose a stance, research for 5 minutes while Groq and DuckDuckGo build a research brief in parallel, then debate live by voice with a talking animated SVG avatar.
Stars
6
7-day growth
No data
Forks
0
Open issues
0
License
No data
Last updated
2026-07-30
AI repository intelligence
FR-AI / ANALYSISWhy it is worth attention
It shows how to build a fairly complete voice-based AI debate experience using only a free Groq API key, a no-key DuckDuckGo search library, browser built-in speech APIs, and a custom SVG avatar, with a simple FastAPI backend and a no-build static frontend.
Who it is for
- Developers prototyping AI voice or debate applications
- Hackathon participants looking for a free-stack AI project
- Students or learners who want to practice argumentation and debate
- Builders evaluating FastAPI, Groq, and browser speech API integration
Use cases
- Practicing spoken debate arguments against an AI opponent
- Prototyping a voice AI agent with parallel research and live rebuttal
- Demonstrating a no-paid-API voice interaction stack
- Learning how to deploy a FastAPI backend and a static frontend separately
Strengths
- No paid APIs required: uses Groq's free tier, DuckDuckGo search without a key, and the browser's built-in Web Speech API and speechSynthesis
- The AI precomputes a research brief with key arguments, likely opponent points, and pre-drafted rebuttals, so it is not reasoning from scratch mid-debate
- Simple architecture: FastAPI backend plus plain HTML/JS/CSS frontend with no build step
- The built-in scorecard reviews the full transcript, flags weak points, and gives one improvement tip
Considerations
- Push-to-talk only; there is no streaming STT, barge-in, or interruption handling yet
- Latency is higher than a production build because the research brief is the only precompute step
- The session store is in-memory and avatar lip-sync is randomized, so it is not multi-user or viseme-accurate
README quick start
The Floor — AI Debater (MVP)
https://ai-debater-seven.vercel.app/ check this out
A local, no-paid-API-required AI debate opponent: pick a category, spin for a topic, take a stance, research for 5 minutes while the AI researches in parallel, then debate live by voice with a talking avatar.
What it uses (all free)
- LLM: Groq — free API key, generous free tier, very fast inference (needed for real-time debate to feel natural).
- Web search: DuckDuckGo via
duckduckgo-search— no key required. - Speech-to-text: your browser's built-in Web Speech API (Chrome/Edge). No key, no server round-trip.
- Text-to-speech: your browser's built-in
speechSynthesis. No key. - Avatar: a custom animated SVG face (no avatar API), mouth-synced to when the AI is speaking.
1. Get a free Groq API key
- Go to https://console.groq.com and sign up (free).
- Create an API key.
- In
backend/, create a file called.env:
GROQ_API_KEY=your_key_here
(Optional: set GROQ_MODEL=llama-3.3-70b-versatile or another Groq-hosted
model — see https://console.groq.com/docs/models for current free models.)
2. Run the backend
cd backend
python3 -m venv venv
source venv/bin/activate # Windows: venv\Scripts\activate
pip install -r requirements.txt
uvicorn main:app --reload --port 8000
Check it's alive: open http://localhost:8000/api/health — you should see
"groq_key_set": true.
3. Run the frontend
The frontend is plain HTML/JS, no build step. Simplest option:
cd frontend
python3 -m http.server 5500
Then open http://localhost:5500 in Chrome (Web Speech API support is best there). Allow camera access when prompted.
How it works
- Welcome → category → wheel: topics come from a curated bank per
category in
backend/topics.py(edit freely to add your own). - Stance: you pick for/against, the AI automatically takes the other side.
- Research (5 min, adjustable): on stance selection, the backend fires
off
build_research_brief()— DuckDuckGo search for both sides of the topic, summarized by Groq into a compact brief (key arguments, likely opponent points, pre-drafted rebuttals). This brief is what the AI draws on live, so it isn't reasoning from scratch mid-debate. Meanwhile you get a notes panel to do your own research in. - **Live
Related repositories
Similar projects matched by category, topics, and programming language.
slvDev
Featured
esp32-ai
A 28.9 million parameter language model runs on an $8 ESP32-S3 microcontroller entirely on-device, generating simple stories at about 9.5 tokens per second.
gavamedia
Featured
deltafin
Deltafin is a research project that runs the 2.8-trillion-parameter Mixture-of-Experts model Kimi K3 on a single Apple Silicon Mac (e.g., M1 Max with 64 GB) at about 16 seconds per token, using exact, reproducible inference with local or streaming expert loading.