FirstModel — From dataset to first model in minutes.

This is a Next.js app (App Router) using Convex for realtime/storage and a Modal‑hosted FastAPI backend for preprocessing, profiling and training. It turns a CSV into a reproducible baseline model and pipeline artifact quickly.

Environment Setup (Convex + Modal)

Set these in the correct places to enable preprocessing:

• Convex (deployment env variables)

  • MODAL_PREPROCESS_URL: The public URL of your Modal FastAPI app endpoint. You can set the base (e.g. https://<app>.modal.run) and the system will append /preprocess automatically.
  • PREPROCESS_WEBHOOK_SECRET: A random secret (e.g., from openssl rand -hex 32) used to authenticate callbacks into Convex.
  • CONVEX_SITE_URL (preferred): Your Convex HTTP Actions base URL (shown in the Convex panel as “HTTP Actions URL”). This is what our callbacks hit.
    • If you prefer, CONVEX_URL or NEXT_PUBLIC_CONVEX_URL can be used as fallback, but CONVEX_SITE_URL is recommended.
  • CLI examples:
    • npx convex env set MODAL_PREPROCESS_URL https://<app>.modal.run
    • npx convex env set PREPROCESS_WEBHOOK_SECRET <your-secret>
    • npx convex env set CONVEX_URL https://<project>.convex.cloud

• Modal (secrets)

  • Create a secret named tabular-ml-agent-backend with key NEXT_PUBLIC_CONVEX_URL (or CONVEX_URL) set to your Convex HTTP Actions URL — the same as CONVEX_SITE_URL above.
  • The backend reads CONVEX_URL and falls back to NEXT_PUBLIC_CONVEX_URL.

• Local (optional, for modal serve)

  • You can export CONVEX_URL in your shell so the backend picks it up when serving locally:
    • export CONVEX_URL="$(grep NEXT_PUBLIC_CONVEX_URL .env.local | cut -d= -f2)"

Health check: Once served/deployed, the Modal app exposes GET /health returning { "ok": true }.

Notes

• Uploads are limited to 1MB in the demo UI (see src/app/studio/UploadClient.tsx).
• Preview defaults to 5 rows and can be adjusted via the row selector.
• “Next” actions are primary; disabled buttons surface tooltips explaining why.
• Results show balanced accuracy plus accuracy/precision/recall/f1 for classification and MAE for regression.

Getting Started

First, run the development server:

npm run dev
# or
yarn dev
# or
pnpm dev
# or
bun dev

Open http://localhost:3000 with your browser to see the result.

You can start editing the landing page by modifying src/app/page.tsx. The page auto‑updates as you edit the file.

This project uses next/font to automatically optimize and load Geist, a new font family for Vercel.

Modal backend changes (important)

• Predict endpoint: skops allow‑list expanded to load common sklearn internals. You can opt‑in to a permissive load for testing via ALLOW_UNTRUSTED_MODELS=1 in the container env.
• Training endpoint: returns additional classification metrics: accuracy, precision (macro), recall (macro), and f1 (macro), each with *_std, in addition to balanced_accuracy. Regression still returns mae and mae_std.

Brand & Metadata

• Brand: FirstModel
• Tagline: “From dataset to first model in minutes.”
• Metadata updated in src/app/layout.tsx (OpenGraph/Twitter, authors, creator/publisher).