How tanishvenpure/ticket-triage-transformer works
This is a support-ticket triage product: a user submits ticket text through the Streamlit chat/dashboard or directly to the FastAPI API; the API validates it, obtains category and urgency predictions from the ONNX inference service, then applies transparent routing rules to return priority, action, required skills, and a matched agent. A separate model-development branch cleans ticket data, evaluates a baseline, fine-tunes a multi-task transformer, and prepares and calibrates the deployed model. The README documents these product stages; unsampled implementation details and external deployment wiring are not inferred.
Main entry points: main.py FastAPI endpoints · models Model artifacts · train.py Fine-tuning loop · inference.py ONNX inference
Components 17
Ticket interface
- Chat and dashboard
dashboard/app.py
Triage service
- FastAPI endpoints
app/main.py - ONNX inference
app/inference.py - Routing decisions
app/routing.py - Triage response
app/main.py
Model development
- Ticket cleaning
src/load_and_clean.py - Clean ticket data
data/processed/tickets_clean.csv - Baseline evaluation
src/baseline.py - Training dataset
src/dataset.py - Fine-tuning loop
src/train.py - Multi-task classifier
src/model.py
Serving preparation
- ONNX export and benchmark
src/quantize_and_benchmark.py - Quantization accuracy check
src/check_quantized_accuracy.py - Temperature calibration
src/calibrate.py - Model artifacts
models - Calibration settings
results/calibration.json
Other components
- Ticket submitter
Connections 20
- Ticket submitter → Chat and dashboard: submits ticket
readme.md - Chat and dashboard → FastAPI endpoints: requests prediction
dashboard/app.py - Ticket submitter → FastAPI endpoints: posts ticket
readme.md - FastAPI endpoints → ONNX inference: classifies text
app/main.py - FastAPI endpoints → Routing decisions: gets routing
app/main.py - FastAPI endpoints → Triage response: returns result
app/main.py - ONNX inference → Model artifacts: loads model and labels
app/inference.py - ONNX inference → Calibration settings: loads temperatures
app/inference.py - Routing decisions → Triage response: adds decisions
app/routing.py - Ticket cleaning → Clean ticket data: writes cleaned data
readme.md - Clean ticket data → Baseline evaluation: evaluates baseline
readme.md - Clean ticket data → Fine-tuning loop: trains on data
- Fine-tuning loop → Training dataset: uses batches
src/train.py - Fine-tuning loop → Multi-task classifier: trains model
src/train.py - Fine-tuning loop → Model artifacts: saves checkpoint
- Model artifacts → ONNX export and benchmark: exports ONNX
readme.md - ONNX export and benchmark → Model artifacts: writes ONNX
src/quantize_and_benchmark.py - ONNX export and benchmark → Quantization accuracy check: checks accuracy
readme.md - Model artifacts → Temperature calibration: calibrates predictions
readme.md - Temperature calibration → Calibration settings: writes temperatures
src/calibrate.py