Adam Rahman@msradam

Riprap turns any NYC address into a citation-grounded flood-exposure briefing in about two minutes. Every claim points back to a [doc_id] in public-record data. No vendor scores, no opaque aggregators. The problem: NYC has spent decades publishing the inputs (Sandy 2012, DEP stormwater, FloodNet, NOAA tides, USGS LiDAR, 311, MTA, NYCHA, schools, hospitals). The data is public. None of it composes itself. Every engineer reassembles the same evidence by hand, per address, from a dozen agencies. Riprap composes it. Three NYC-specialised foundation-model fine-tunes shipped Apache 2.0, trained on a single AMD MI300X via AMD Developer Cloud: TerraMind-NYC-Adapters: LULC mIoU 0.5866 (huggingface.co/msradam/TerraMind-NYC-Adapters) Prithvi-EO-2.0-NYC-Pluvial: flood IoU 0.5979 vs 0.10 on Sen1Floods11 base, 6x lift (huggingface.co/msradam/Prithvi-EO-2.0-NYC-Pluvial) Granite-TTM-r2-Battery-Surge: NYC Battery surge nowcast, MAE 0.109 m, 41% better than persistence (huggingface.co/msradam/Granite-TTM-r2-Battery-Surge) About 25 atomic data probes fan out, organised as the Five Stones: Cornerstone (hazard memory), Keystone (asset registers), Touchstone (live state), Lodestone (forecasts), Capstone (citation-grounded synthesis). The Five Stones taxonomy is city-agnostic. Only the probes plugged into each Stone change. To port Riprap to another flood-vulnerable region, you reimplement each Stone's collect() against local data and retrain the EO and time-series fine-tunes on the jurisdiction's imagery and tide gauges. The agentic shell stays the same. The pipeline is a Burr FSM with one @action per probe; Mellea rejection sampling re-rolls until four grounding checks pass. Inference runs through a LiteLLM Router targeting either AMD MI300X (vLLM, BF16 Granite 4.1 8B) or NVIDIA L4 (vLLM, FP8 Granite 4.1 8B) co-resident with the EO model stack. Full architecture, methodology, and licence map: github.com/msradam/riprap-nyc

Grab your ideas, spin up compute, and start building what’s next! Start with $100 in AMD Developer Cloud credits and begin building right away. Get access to hands-on tutorials, workshops, and curated learning resources to help you build faster and go further. To unlock credits and join the hackathon, sign up for the AMD AI Developer Program. 🤝 Go solo or team up — ship something real. 💰 $21,500+ total prize pool & 1 × AMD Radeon™ AI PRO R9700 GPU up for grabs. 📍 On-site Venue (May 9-10): MindsDB SF AI Collective 3154 17th St, San Francisco, California, USA 📍On-site participation is by invitation only. Travel and accommodation expenses will not be covered. 🧑‍💻 Developers, builders, and innovators - bring your boldest ideas and let’s turn them into real AI systems on AMD.