Explain how the system compares Weft-style orchestration vs traditional Python pipelines.
Demonstrate that modular orchestration reduces token usage and cost versus linear context passing.
- frontend/: UI
- backend/: server + pipelines
- backend/pipelines/:
- weft_pipeline.weft (conceptual)
- baseline_pipeline.py (linear)
- data/: sources
- configs/: pricing and demo configs
-
Weft-style:
- Retrieval → summarization → synthesis
- Passes compact structured outputs between steps
-
Python baseline:
- Single linear flow
- Re-sends entire context buffer at each step
Token cost grows with context size. Weft reduces cost by controlling what data flows between steps.
- Same inputs across all modes
- Centralized telemetry for fair comparison
- Deterministic runs for reproducibility
- Telemetry + pricing
- Baseline pipeline
- Weft-style pipeline
- Backend routing
- UI comparison