Post
159
Riprap: citation-grounded NYC flood-exposure briefings π
Any NYC address β a four-section, citation-grounded flood-exposure briefing in about two minutes. Every claim points back to a
π¬ Live demo: lablab-ai-amd-developer-hackathon/riprap-nyc
About 25 atomic data probes fan out across NYC datasets, Sentinel-2 imagery, live sensors, and forecasts, organized as the Five Stones:
πͺ¨ Cornerstone : hazard memory
ποΈ Keystone : asset registers
π‘ Touchstone : live state
π§ Lodestone : forecasts
βοΈ Capstone : citation-grounded synthesis (Granite 4.1 8B + Mellea rejection sampling, four grounding checks per draft)
Three NYC-specialised foundation-model fine-tunes shipped Apache 2.0 alongside, trained on a single AMD MI300X via AMD Developer Cloud:
π°οΈ msradam/TerraMind-NYC-Adapters ( msradam/TerraMind-NYC-Adapters) : LULC mIoU 0.5866, +6.13 pp over full-FT baseline (plus Buildings + TiM heads).
π msradam/Prithvi-EO-2.0-NYC-Pluvial : ( msradam/Prithvi-EO-2.0-NYC-Pluvial) : flood IoU 0.5979 vs 0.10 base, a 6Γ lift.
π msradam/Granite-TTM-r2-Battery-Surge ( msradam/Granite-TTM-r2-Battery-Surge) : Battery surge nowcast, MAE 0.1091 m, 41% better than persistence.
Repo: https://github.com/msradam/riprap-nyc
If this is the kind of agentic AI civic tech should be building toward, drop a like on the Space!
The foundation-model teams whose work made this possible: @ibm-granite @ibm-nasa-geospatial @ibm-esa-geospatial @amd
#agenticai #civictech #climateai #floodresilience #nyc #foundationmodels #granite #terramind #prithvi #amd #mi300x
Any NYC address β a four-section, citation-grounded flood-exposure briefing in about two minutes. Every claim points back to a
[doc_id] in public-record data.π¬ Live demo: lablab-ai-amd-developer-hackathon/riprap-nyc
About 25 atomic data probes fan out across NYC datasets, Sentinel-2 imagery, live sensors, and forecasts, organized as the Five Stones:
πͺ¨ Cornerstone : hazard memory
ποΈ Keystone : asset registers
π‘ Touchstone : live state
π§ Lodestone : forecasts
βοΈ Capstone : citation-grounded synthesis (Granite 4.1 8B + Mellea rejection sampling, four grounding checks per draft)
Three NYC-specialised foundation-model fine-tunes shipped Apache 2.0 alongside, trained on a single AMD MI300X via AMD Developer Cloud:
π°οΈ msradam/TerraMind-NYC-Adapters ( msradam/TerraMind-NYC-Adapters) : LULC mIoU 0.5866, +6.13 pp over full-FT baseline (plus Buildings + TiM heads).
π msradam/Prithvi-EO-2.0-NYC-Pluvial : ( msradam/Prithvi-EO-2.0-NYC-Pluvial) : flood IoU 0.5979 vs 0.10 base, a 6Γ lift.
π msradam/Granite-TTM-r2-Battery-Surge ( msradam/Granite-TTM-r2-Battery-Surge) : Battery surge nowcast, MAE 0.1091 m, 41% better than persistence.
Repo: https://github.com/msradam/riprap-nyc
If this is the kind of agentic AI civic tech should be building toward, drop a like on the Space!
The foundation-model teams whose work made this possible: @ibm-granite @ibm-nasa-geospatial @ibm-esa-geospatial @amd
#agenticai #civictech #climateai #floodresilience #nyc #foundationmodels #granite #terramind #prithvi #amd #mi300x