Why should you care?

AIPredict ShopNest mobile apps call a .NET API, not an embedded model on the device.

See it live — copy this example

Use a .NET console or Web API project with Microsoft.ML. Run dotnet run after pasting.

app.MapPost("/api/recs/products", (RecRequest req, PredictionEnginePool<UserProduct, ProductScore> pool) =>
{
    var scores = req.CandidateIds.Select(id =>
        pool.Predict(new UserProduct { UserId = req.UserId, ProductId = id }).Score);
    return Results.Ok(scores.Zip(req.CandidateIds).OrderByDescending(x => x.First).Take(req.TopN));
});

What happened?

• Use PredictionEnginePool for thread-safe scoring.
• Accept candidate ids from search, score, return ordered ids.

Practice next

1. Register pool in Program.cs.
2. MapPost rec endpoint.
3. Return id+score JSON.
4. Add cache key per user.
5. Return explain metadata (category match).

Remember

Pool for inference. Candidates in, ranked out. Stable DTO contract.

AIPredict rec API

App requests top-10 from search candidates.

Outcome: Sub-50ms ranked lists at scale.