AI Agent System
An LLM-driven NBA player recommendation and scouting workflow built around team need diagnosis, player strength vectors, fit ranking, robustness checks, and grounded Q&A.
Pipeline nodes
11
WebApp registry
Eval metrics
7
zero-shot vs tools
Backend
FastAPI
imports nba_agent
overview
The backend imports the original NBA-Roster-Upgrade-Agent-Webapp package from the v2-webapp-agent branch. It calls the original parser, roster agent, agentic tool selector, need reasoning, player-strength builder, fit ranker, sensitivity checker, grounded Q&A, zero-shot evaluator, benchmark helpers, and radar renderer without importing Streamlit.
role
Original project authoring and portfolio integration: replaced the simplified portfolio wrapper with a FastAPI bridge around the WebApp computation modules and a Next.js agent console.
full WebApp backend wrapper
Calls /api/nba/run on https://ruizelab-api.onrender.com and displays the original WebApp pipeline, traces, evaluation, and grounded Q&A.
run controls
data contract
Requires teams.csv, games.csv, and games_details.csv. Missing files return a clear error without recorded fallback.
natural-language query input
model status
Model openrouter/free. Deterministic mode.
how to read this result
Selected tools show what the planner surfaced first. Closed cards can still be opened because the backend returns the full AgentResult trace for audit.
Configure the WebApp controls and run NBA Roster Upgrade Agent to display the original pipeline output.
architecture flow
The live pipeline trace appears in the backend runner after execution. This section shows the original project components that the backend wraps.
planner
Converts the natural-language roster request into team, goal, top-k, recent-game window, and availability filters.
planner
Selects visible workflow sections, validates dependencies, records skipped tools, rationales, and fallback state.
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.
ranker
Ranks players by fit score, generates recommendation cards, and checks stability under need-weight perturbation.
tools and models
planner
Converts the natural-language roster request into team, goal, top-k, recent-game window, and availability filters.
planner
Selects visible workflow sections, validates dependencies, records skipped tools, rationales, and fallback state.
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.
ranker
Ranks players by fit score, generates recommendation cards, and checks stability under need-weight perturbation.
example input
Recommend top 5 players for the Warriors to improve interior defense using the last 10 games. Only include players with at least 20 games and 18 average minutes.
final result
The FastAPI wrapper returns the original WebApp AgentResult structure: parsed query, selected/skipped tools, Tool A/B/C tables, sensitivity output, scouting summary, grounded Q&A, zero-shot comparison metrics, benchmark rows, and radar SVGs.
limitations