Roadmap

GridMind is a working prediction pipeline, not a finished product. Here's what's shipping next, the boundaries it's deliberately shipping around today, and an analysis layer that's already designed but not yet built.

What's next

  1. 1

    More training data

    Ingest 2–3 seasons instead of one, for a less anemic training set.

  2. 2

    Tyre & strategy features

    Add the stints endpoint — tyre compound and strategy features.

  3. 3

    AutoML upgrade path

    Try ML.NET's AutoML API once the SDCA baseline is trusted.

  4. 4

    Background ingestion

    Move ingestion onto a background queue so the endpoint returns 202 immediately.

  5. 5

    Retire the legacy import

    Remove the one-off SQLite importer used to migrate the original console prototype.

Known scope cuts

Deliberate boundaries for a first release, not oversights.

  • Single season of data

    ~480 driver-race rows — enough to demonstrate the pipeline end to end, not enough for the predictions to be authoritative.

  • No tyre or strategy features

    The stints endpoint isn't ingested yet.

  • Qualifying gap sentinel

    A missing qualifying lap gets a 99 sentinel value rather than proper imputation.

  • Synchronous ingestion

    Ingestion runs inside the request. A background queue is a later concern.

  • Legacy SQLite import

    Exists only to migrate the original prototype's data — development-only, slated for removal.

Planned: an AI analysis layer

Designed, not yet built. Two approved specs add interpretation beside the prediction pipeline, never inside it — the classifiers and the Monte Carlo simulator stay the only source of numbers.

Explanations & analyst chat

Coefficient-grounded, per-feature explanations for a prediction; a generated race-preview narrative; and a tool-calling analyst chat that answers questions by calling GridMind's own use cases — running on a local Ollama model.

Hosted provider & semantic search

A hosted OpenAI provider for chat and embeddings in production, explanations extended to races that have already run, a background job that keeps previews fresh, and semantic search over past races via embedded fact sheets in Postgres.

Follow along in the dashboard.

Open the dashboard