Work Done
Inspected mono dashboard-api dependencies and histo routes, assistant execution, and dashboard-frontend/shared TileViewer source. No application changes. The HTTP API can move to TypeScript, but current TIFF tiling uses large_image and vxAgent runs Python scripts using vxdata.sdk, NumPy, Polars, matplotlib and sklearn. These capabilities can remain separate Python workers/services. The shared viewer already consumes Python HTTP endpoints; switching backend language alone does not address viewer issues.
Pitfalls
The dashboard pyproject.toml is not the full runtime dependency inventory: pixi.toml declares imaging and assistant dependencies. The histo tile-serving SOP predates the browser GeoTIFF/S3 path and mentions a 1Gi limit; current histo.py comments describe a 4Gi limit and bounded cache. Do not treat those historical limits as verified deployment state.
Improvements
Assess file formats and transports separately from API language. Current dashboard rejects S3 Zarr, caches expiring presigned URLs for the session, and uses one source type for base image/overlays. Shared Zarr reader expects v3 and three c/y/x axes. Validate representative real slides, overlays, range requests and auth before replacing TIFF tiling. No production issue was reproduced in this assessment.