A young field, watched every day.
“To what extent are generative AI and no-code tools levers of professionalisation and organisational efficiency for a sports association with limited human and financial resources?”
A young topic
Research on generative AI and no-code tools in amateur sport is only starting. The few relevant sources have to be caught as soon as they appear.
Three databases
Google Scholar, OpenAlex and Cairn cover international and French-language research. Checking them by hand every day would eat the time meant for writing.
The same constraint
One person and close to zero budget: the engine runs under the very constraints the thesis studies.
A research assistant that never sleeps.
Every dot below is one candidate source. Follow it through the scan, the local model and the vault, or send your own source through the pipeline.
- Candidate
- Kept → vault
- Escalated → Sonnet
- Dropped
Simulation at the pipeline’s real rates: about 90 candidates a day over three passes, roughly 5% kept. The escalation frequency is illustrative.
Hover, tap or tab through any node.
- 01 Scan three databases, three times a day
- 02 Triage locally with a fine-tuned Qwen3 4B
- 03 Escalate complex analyses to Claude Sonnet 5.5
- 04 Write notes, canvases and bibliography to the vault
Ninety in, a handful out.
One simulated day at the real rates. Each square is a candidate: the three rows are the three daily scans, then the local model keeps what serves the question.
- Kept → vault
- Dropped at triage
Small model on duty, big model on call.
The local model does the daily work for free. Claude only steps in when an analysis needs more depth, which keeps the bill at a few cents a month.
Qwen3 4B
- Fine-tuned with LoRA
- Grounded with RAG on reference documents
- Served by Ollama on my machine, 24/7
- Reads every candidate, sorts, selects and summarises
→ Stays local on Qwen3 4B
Drag to see the routing rule. Illustration.
Claude Sonnet 5.5
- Called only when the analysis is complex
- Deeper explanations of a source or a debate
- Visuals for the vault
Where it all lands.
The vault is organised around the concepts of the research question. Kept sources arrive as summarised notes, complex ones as deeper analyses, and the bibliography follows on its own.
Generative AI & no-code as levers for sports associations with limited resources.
Nine pieces, one routine.
The engine is its own field test.
One person, almost no budget, and a research routine that scans, sorts, summarises and files every day. A small-scale test of the thesis question: what generative AI can do when time and money are scarce.
- Status
- Running since October 2026
- Origin
- Adapted from my own AI-news watch pipeline
- Cost
- ≈ €0 a month locally, a few cents of Sonnet
- Code
- Private repository while the thesis is in progress
- Next
- Available soon for other teams, adaptable to any sector