Open-source football AI trained on 800 analyzed matches, 44 original metrics, and 370+ tactical sources. From grassroots coach to professional analyst.
Live platform: llm.fintalab.com
Sports ecosystem: fintalab.com
Model: Hugging Face
Finta LLM democratizes football intelligence across all levels of the game. It wraps a 7-billion-parameter language model — fine-tuned exclusively on football knowledge — inside a Jupyter-style notebook interface with an interactive knowledge map. The model translates sports science research into practical coaching responses, personalized by role.
No hallucinated statistics. No speculative data. Every answer is grounded in the training corpus and explicit about its limits.
| Property | Value |
|---|---|
| Architecture | 7B parameter fine-tuned LLM |
| Analyzed matches | 800 |
| Original metrics | 44 |
| Statistical fields | 350+ |
| Tactical sources | 370+ |
| Hosting | Hugging Face (free access) |
| Languages | English · French |
The model personalizes responses based on the user's declared role:
| Role | Response focus |
|---|---|
| Coach | Training design, session structure, tactical implementation |
| Analyst | Metric interpretation, data-to-decision translation |
| Sports scientist | Research application, load management, periodisation |
| Player | Technical development, self-assessment, preparation |
A Jupyter-style interface with 14 structured cells:
| Cell | Content |
|---|---|
| 1 | Manifesto — mission, scope, limitations |
| 2–4 | Use cases by role |
| 5–7 | Knowledge domain overview |
| 8–10 | Architecture and methodology |
| 11–12 | Metric definitions |
| 13–14 | Usage guides and examples |
Accessible at /knowledge_map — an interactive visual graph of:
- 460+ interconnected concepts
- 11 knowledge domains
- RAG + LLM structure for concept identification
- Visual navigation across tactical, physical, psychological, and analytical football domains
Switch between the notebook and knowledge map via the topbar.
The platform is built on four non-negotiable commitments:
No hallucinated data — The model never invents statistics, match results, or player data. It acknowledges uncertainty explicitly.
Source grounding — Tactical claims are traced to the training corpus. Responses distinguish between established research and contextual inference.
Youth safety — Specific safeguards apply to content involving youth athletes. The model declines to provide protocols outside appropriate professional contexts.
Honest limitations — The system states what it cannot reliably answer, rather than filling gaps with plausible-sounding fiction.
| Route | Serves |
|---|---|
/ |
Main notebook interface (html/index.html) |
/knowledge_map |
Interactive knowledge map (html/mindmap.html) |
/* |
Static assets (CSS, JS, images, fonts) |
.
├── server.js # Static file server — routes, MIME types, port 5000
├── html/
│ ├── index.html # Notebook interface — 14 cells, topbar navigation
│ └── mindmap.html # Knowledge map — 460+ concept graph
├── css/ # Stylesheets — notebook, knowledge map, themes
├── js/ # Client-side logic — cell rendering, graph interaction
├── img/ # Icons, diagrams, visual assets
├── package.json # Node.js dependencies
└── .replit # Replit environment configuration
JavaScript · HTML · CSS · Node.js
node server.jsThen open http://localhost:5000. No build step required.
| Resource | URL |
|---|---|
| Live platform | llm.fintalab.com |
| Knowledge map | llm.fintalab.com/knowledge_map |
| Sports platform | fintalab.com |
| Model | Hugging Face |
| GitHub | fintasportscorp-rgb/Finta_llm |