Ruize Ma
A polished portfolio for agent workflows, biomedical reasoning, and predictive modeling, presented as calm product-grade system dashboards backed by the original project code and artifacts.
Featured systems
3
agent, reasoning, prediction
Backend
FastAPI
wraps original project code
Demo data
real
models or original artifacts
system overview
An LLM-driven roster recommendation and scouting workflow built around team need diagnosis, player strength vectors, and fit ranking.
A CURE-Bench Plan-Act-Verify system that uses a model planner, biomedical tools, curated Tool Facts, and a final answer pass.
A backend-wrapped Streamlit project that loads the original saved model artifacts and returns annual cost predictions with uncertainty and feature drivers.
featured work
Each project calls a FastAPI wrapper that either recomputes with the original code path or exposes provenance-backed original artifacts when live dependencies are unavailable.
AI Agent System
An LLM-driven roster recommendation and scouting workflow built around team need diagnosis, player strength vectors, and fit ranking.
Original tools
A-C
need, strength, ranking
Recorded pool
236
eligible players in notebook
Mode
CSV-gated
live with NBA_DATA_PATH
Reasoning System
A CURE-Bench Plan-Act-Verify system that uses a model planner, biomedical tools, curated Tool Facts, and a final answer pass.
Published acc.
0.69564
fine-tuned GPT-4.1 + tools
Passes
2
plan then answer/verify
Tool families
6
FDA, DailyMed, RxNav, more
Predictive Model
A backend-wrapped Streamlit project that loads the original saved model artifacts and returns annual cost predictions with uncertainty and feature drivers.
Inputs
6
age, BMI, children, sex, region, smoker
Best model R2
0.846
XGBoost comparison metric
Backend
FastAPI
wraps original shared.py
build principles
planner
Converts the natural-language roster request into team, goal, top-k, recent-game window, and availability filters.
analytics
Uses recent team statistics and league z-scores to compute a need-weight vector.
features
Builds player strength vectors from box-score features and standardized skill indicators.