BetEdge

Twenty matches at once. Built a sports-prediction tool. The math behind every tip is visible, no more guessing in Excel.

• 01User surface

  The whole experience the user sees

  This frame shows the live product: data-driven bet analyser · parlay suggestions, best odds, probability math, not guesswork. Every component is ours · scope, design, code, deploy.

• 02Stack behind the screen

  What's powering it: Flask, FastAPI, SQLite

  4 stack components run behind this frame · Flask, FastAPI, SQLite drive the visible UI; the rest sit in the data layer. All studio-owned.

• 03What we shipped

  Match and parlay analysis in one place

  Data-backed betting decisions

• 04Status

  Private deploy · under NDA.

  Per the client's request the URL stays private · the build, architecture, and lessons can be shared in a scoping call.

• 01 · BRIEF

  Why random picks lose long-term.

  Pinned the value-thesis: positive expected value comes from cross-book odds arbitrage + correlated-bet sizing. Spec'd a tool that ranks parlays by EV, not by 'looks good'.

• 02 · ARCHITECTURE

  Stack decisions before any code.

  Decision doc captured the data flow, Flask, FastAPI, SQLite, Python role split, and the failure modes we'd handle in v1 vs defer. Cross-service boundaries (where AI ends and the web app begins) were drawn here so neither side leaked into the other later.

• 02 · BUILD

  Flask front + FastAPI worker + SQLite store.

  Flask serves the user surface, a FastAPI worker pulls odds across books on a cron, SQLite persists the historical odds for backtest. Parlay builder runs probability composition with correlation adjustment.

• 04 · POLISH

  Performance, accessibility, and observability.

  PSI / a11y / coverage budgets enforced as launch gates. Logging + metrics wired before cut-over · the team can answer 'is it working?' from a dashboard, not a Slack thread. Threat-model checklist signed off before traffic hits the box.

• 03 · SHIP

  Web + API · explanations on by default.

  Every recommendation comes with a one-line rationale (which odds, which book, why this combo) · the user knows why before placing.

• 01Ingest

  Cross-book odds crawler

  Pulls odds from multiple books on a cron · normalises to common markets before comparison.

• 02Parlay

  Probability-composed parlays

  Builder evaluates correlation between legs · doesn't pretend independent events stay independent.

• 03Best odds

  Per-outcome book pick

  For each leg, the recommendation surfaces the book with the best odds · easy to follow at the time of placement.

• 04Explainer

  One-line rationale per pick

  Why this combination, why this book, what the EV looks like · transparent default.

• 
  01Frame

  Elo Ratings · empty state, builds itself

  Empty-state honesty · 'No Elo data yet. Ratings are built automatically as games complete. Run Refresh to start populating.' No fake placeholder rows, no demo data masquerading as real.

• 
  02Frame

  Best Bets · live analysis + confidence-tiered picks

  'Run Analysis' button polls for results in real time. Four KPI tiles (Games Analysed 2, Value Bets 3, High Confidence 1, Medium Confidence 2) sit above the Top 5 Bets of the Day · a visitor sees the funnel without scrolling.

• 
  03Frame

  Value bet · 4-cell maths visible

  Per-bet card opens up the maths: Edge 161.2%, our model probability 22.5%, the book's implied 8.6%, Kelly fraction 7.6% · plus form context. The user sees the value calculation, not just a recommendation.

• 
  04Frame

  Parlays · Bold (103×) + Longshot (295×) built by model

  Two pre-built parlays scoped to risk appetite: Bold (103.78× · 1.3% combined probability) for the disciplined bettor, Longshot (295.11× · 0.9% combined probability) for the lottery-ticket player · stake-to-payout shown for each.