How I Built It

The build started from the Heat Index formula, since that's the number the

whole app orbits around: avg_passion * 0.5 + avg_sentiment_intensity * 3 + log2(take_count + 1) * 2. Volume matters (a rivalry with one take isn't

"hot"), but so does how emotionally loaded the language is — and that's

where Google AI comes in.

Gemini reads each take and classifies its sentiment:

prompt = (

    "Classify the overall emotional sentiment of this football fan "

    "comment as exactly one word — positive, negative, mixed, or "

    f"neutral. Reply with only that one word.\n\nComment: {text}"

)

response = client.models.generate_content(model="gemini-2.5-flash", contents=prompt)

That categorical result gets mapped to a numeric intensity — fury counts

exactly as much as joy, both are passion — so it drops straight into the

Heat Index math.

For the fun part, Gemini also turns the most recent takes for a rivalry into

a punchy one-liner:

prompt = (

    "You are a stadium hype announcer. In under 40 words, deliver a "

    f"punchy verdict on the {team_a_name} vs {team_b_name} World Cup "

    f"rivalry based on these fan takes: {joined}"

)

Snowflake handles the other half of the job: storing every take and

computing the leaderboards with real SQL — aggregation, a derived metric,

and a RANK() window function per rivalry and per fanbase. It's a clean

split: Gemini reads the emotion, Snowflake turns it into a ranking.

The backend is a small FastAPI service with two independent fallbacks,

keeping the whole flow explorable without handing out API keys for a

weekend project: no GEMINI_API_KEY → sentiment scoring falls back to a

keyword heuristic; no SNOWFLAKE_ACCOUNT → the whole API runs in demo mode

with seed data.

The frontend leaned into the subject: a split-flap "departure board" digit

animation for the Heat Index, a scrolling terrace-chant ticker, and a

submission form styled like a stadium chalkboard — an attempt to make the

data feel like the thing it's measuring.

Prize Categories

Submitting for Best Use of Google AI and Best Use of Snowflake — Gemini does the real intelligence work in this project: reading the emotion behind every fan take and writing the Hype Verdict. Snowflake plays an honest supporting role as the data warehouse, storing every take and doing the ranking analytics that turn Gemini's scores into a live leaderboard.

Thank you.