◵
Physics-Informed ML
Neural operators and gray-box ODE models that respect the physics. Ported a per-well groundwater model to NVIDIA PhysicsNeMo, benchmarked on GPU, and tested it where it is hardest: wells the model has never seen.
Selected work
Projects that show the number, not just claim it.
Flagship · NVIDIA stack
Python · PyTorch · PhysicsNeMo · CUDA
HydroPhysicsAI
GPU physics-informed neural operators for groundwater: one operator trained across all
61 wells, benchmarked head-to-head against the per-well gray-box ODEs. A real PhysicsNeMo
port with .mdlus checkpointing, a
CUDA GPU benchmark, leave-one-well-out generalization, and a probabilistic forecast model
with calibrated intervals. Honest by design: it openly reports where it trails the gray-box
baseline.
- 0.236 → 0.565 KGE on unseen wells (beats climatology 0.446)
- 14× faster on RTX 4070 SUPER vs CPU (bf16 AMP)
- Live Gradio demo, model card, technical write-up, CI
Open source · ★7
Python · Streamlit · PyPI
AquaScope
An open-source toolkit for water data, hydrology, and agricultural water management. 15 unified data sources (USGS, FAO, GEMStat, EU WFD and more), Bulletin 17C flood frequency, FAO-56 crop water, baseflow separation, and an agentic AI engine that scores 26 methodologies and auto-executes 7 pipelines. CAMELS-validated, 500+ tests, on PyPI.
- 15 global water-data collectors behind one API
- Agentic recommender: OpenAI / Groq / Hugging Face / Ollama
- Streamlit dashboard, 500+ tests, reproducible demos
You
→
Agent layer · LLM
collect · route · explain
collect · route · explain
⌄ validation gate · rejects bad input ⌄
TRUST ZONE — pure · tested · cited
coefficientsmass balancesizingoptimizer
→ sized system · bill of materials · operating envelope · cited sources
Agentic AI · active
Python · LangChain · RAG + tool-calling · Ollama / NVIDIA / HF
Agronaut
A personal agronomy agent for aquaponics. Instead of retrieving what a paper said, Agronaut computes the answer for your specific system: a tool-calling LLM collects facts and routes, while a deterministic, fully tested engineering core does the math, now calibrated against an open real-pond dataset. It remembers your system across sessions and reasons over a curated troubleshooting knowledge base. Every result lists the coefficients it used (with sources) and what it does not model, so a confidently-wrong design can never masquerade as complete.
- Trust-boundary architecture: a tool-calling LLM agent over a verifiable, cited engine
- Cross-session memory plus a deep troubleshooting knowledge base (DO, nitrogen cycle, pH, nutrient deficiencies, failures)
- Calibrated against published reference systems and an open real-pond dataset
- Runs as a Streamlit app or a Telegram agent; pluggable Ollama · NVIDIA · HF; design & optimize need no LLM
- Underlying research granted Taiwan Utility Model Patent M661364 (2024)
claude > /paper-agent review manuscript.docx ✓ runtime Claude Code plugin v1.5.0 ✓ citations Semantic Scholar MCP ✓ guard anti-fabrication: 0 invented refs → manuscript.reviewed.docx
Claude Code plugin · ★6
Claude Code plugin · Semantic Scholar MCP · .docx
paper-agent
A Claude Code plugin that turns the agent into a disciplined manuscript
collaborator: five modes (draft, review, revise, proofread, audit), citations resolved
through the Semantic Scholar MCP (auto-wired on install), and hard
anti-fabrication guardrails with pause-and-confirm after every section. Calibrated for
hydrology and IEEE journals, with a generic profile for any quantitative-science field and a
clean .docx round-trip.
- Installable Claude Code plugin (v1.5.0) with bundled try-it demos
- Auto-wires the Semantic Scholar MCP for real, non-fabricated citations (no API key needed)
- Hydrology + IEEE journals, plus a generic quantitative-science profile
Install in Claude Code
/plugin marketplace add Rekin226/paper-agent
/plugin install paper-agent@paper-agent