FilmRisk.AI scores a film's greenlight risk from 2,200+ Indian films by mixing a gradient-boosted model with Bayesian priors. Prior pulls thin-data predictions back to base rates. We backtest religiously and publish the failure points — because scoring moves money and pretending a model is magic ends careers.
Most side projects end at "an API collected and a chart made." FilmRisk.AI was painstaking because it has a judgment: how risky is a movie $ to greenlight? Wrong scores are a real cost, so we built-in honesty rails.
The 10-component engine
Internal scoring blends 10 feature groups — genre, cast, director, release window, budgets, star momentum — each fed into the model as context. The architecture:
- Gradient-boosted ML learns rich interaction between the 10 components
- Bayesian Priors anchor thin groups: when a 1-film director appears, the score defers toward the category average instead of over-committing
- Backtesting is non-negotiable — runs on the loading data set each change
What honest backtesting surfaced
- Some star-heavy bundles score under the market prior — and the engine says so
- Sequel and date signals beat cast creativity in some timeless categories; the engine shows the effect honestly
- Failure mode: this is a selection engine, not a prediction oracle — for a fan, the score helps pick, it doesn't guarantee a hit
Numbers that matter
- 2,200+ Indian films in training
- 10 components in the score
- Priors that shrink thin data toward category averages
- Backtesting results published on the repo
What I'd do differently
- Ship the API before the blog. The engine's value is a promptable scoring grid; I papered over it for a pretty UI.
- Version the backtest data. Anyone can re-run, but no one can re-fake — pin the sample.
For PMs: this is what measure-before-judge looks like when your product is literally judgment. Pair it with numbers-led product decisioning for the same muscle in SaaS.