1. Problem

Sports content is abundant. The scarce thing is a trustworthy story layer that turns raw events into something editors, sponsors, and product teams can act on without losing factual grounding.

2. Approach

Attimo V0.1 structured the interface around beats, state, arcs, and output variants. That made the narrative layer inspectable instead of mystical.

  • Beats capture the moments worth talking about.
  • State preserves the factual match context around each beat.
  • Arcs and variants let product users inspect different story shapes without hallucinating new facts.

3. Evidence

Narrative timeline
The narrative timeline made the model logic visible enough for product and editorial review.
Caption panel with QA badges
QA badges next to generated output turned trust from a vague claim into something reviewable.

4. Outcome

Attimo proved a product primitive: one event stream can feed many audiences if the narrative layer is explicit, inspectable, and tied back to facts. That is what helped it move from demo to early commercial traction.

5. Tech stack

  • Typed Python domain model for beats, state, arcs, and variants
  • Streamlit UI for rapid product iteration
  • Local LLM inference for fast generation and QA loops

6. Useful links

7. Related reading

8. Call to action

If you are building a narrative product around event data and need the output to stay explainable, I can help with the data model, UI, and QA surface.