How allenwalker3/playcall works
playcall is a natural-language NFL play-by-play analytics application. A user asks a question in the chat UI; the backend builds context for an LLM, dispatches its tool calls, and returns a streamed answer. The main query stages are player lookup, column discovery, and SQL execution against DuckDB views over parquet data. The product also includes data synchronization, provider/model and settings controls, and an optional evaluation workflow. Relationships for unsampled modules are limited to those identified by the source index or README.
Main entry points: api.js API client · app.py Chat API · db.py DuckDB views · App.jsx Chat interface
Components 24
Web interface
- Chat interface
frontend/src/App.jsx - Chat state
frontend/src/hooks/useChat.js - API client
frontend/src/lib/api.js
Chat service
- Chat API
backend/app.py - Prompt builder
backend/schema/system_prompt.py - LLM providers
backend/llm/base.py - Ollama adapter
backend/llm/ollama.py - OpenAI adapter
backend/llm/openai.py - Anthropic adapter
backend/llm/anthropic.py
Query and data
- Tool dispatcher
backend/tools/registry.py - Player search
backend/tools/players.py - Column search
backend/tools/columns.py - SQL execution
backend/tools/sql.py - DuckDB views
backend/db.py - Curated schema
backend/schema/pbp_curated.py
Data and settings
- Data sync UI
frontend/src/components/SyncModal.jsx
Evaluation
- Evaluation UI
frontend/src/components/EvalView.jsx - Evaluation state
frontend/src/hooks/useEval.js - Evaluation API
backend/app.py - Evaluation runner
eval/runner.py - Evaluation cases
eval/cases.py
Other components
- Question asker
- Language model
- NFL parquet data
Connections 32
- Question asker → Chat interface: asks
README.md - Chat interface → Chat state: uses
frontend/src/App.jsx - Chat state → API client: streams chat
frontend/src/hooks/useChat.js - API client → Chat API: posts messages
frontend/src/lib/api.js - Chat API → Prompt builder: builds context
backend/app.py - Chat API → LLM providers: requests completion
backend/app.py - Chat API → Tool dispatcher: dispatches tools
backend/app.py - LLM providers → Ollama adapter: implemented by
backend/llm/ollama.py - LLM providers → OpenAI adapter: implemented by
backend/llm/openai.py - LLM providers → Anthropic adapter: implemented by
backend/llm/anthropic.py - Ollama adapter → Language model: calls
- OpenAI adapter → Language model: calls
- Anthropic adapter → Language model: calls
- Tool dispatcher → Player search: dispatches
backend/tools/registry.py - Tool dispatcher → Column search: dispatches
backend/tools/registry.py - Tool dispatcher → SQL execution: dispatches
backend/tools/registry.py - Player search → DuckDB views: queries
backend/tools/players.py - Column search → DuckDB views: reads columns
backend/tools/columns.py - SQL execution → DuckDB views: executes SQL
backend/tools/sql.py - DuckDB views → NFL parquet data: reads
backend/db.py - Prompt builder → Curated schema: renders schema
backend/schema/system_prompt.py - Prompt builder → DuckDB views: gets columns
backend/schema/system_prompt.py - Chat interface → Data sync UI: opens sync
frontend/src/App.jsx - Data sync UI → API client: requests sync
frontend/src/components/SyncModal.jsx - API client → Chat API: calls data endpoints
- Chat interface → Evaluation UI: shows evaluation
frontend/src/App.jsx - Evaluation UI → Evaluation state: uses
frontend/src/components/EvalView.jsx - Evaluation state → API client: streams evaluation
frontend/src/hooks/useEval.js - API client → Evaluation API: posts evaluation
frontend/src/lib/api.js - Evaluation API → Evaluation runner: runs cases
backend/app.py - Evaluation runner → Evaluation cases: scores
eval/runner.py - Evaluation runner → Chat API: streams chat requests
eval/runner.py