Rotates eight orders I wrote and labeled myself. The intended label prints after the model commits.
View or paste the work order
Paste your own
Reading
Waiting for a work order
0.00 confidence
your text · no intended label to compare
Every category score
Every category's score
Extracted fields
What the rule-based pass pulled out
regex extraction; the LLM extractor is a notebook result, not this endpoint (note 5)
No fields returned.
Similar past cases
Closest past cases from the corpus
No similar cases returned.
Calibration record · LoRA vs full fine-tune vs TF-IDF
| Macro F1 94.4% | DistilBERT + LoRA, 743K trainable params, 1.1% of the model, 2.8 MB artifact |
|---|---|
| 95.9% | full fine-tune, 67M params, 256 MB |
| 93.5% | TF-IDF + logistic regression baseline |
| Failure-mode extraction | 13% rule-based to 70% with the LLM extractor |
| Hybrid extractor | 56% fewer LLM calls but far less than 56% of the uplift kept; regex confidence is a poor escalation gate |
| Held-out test | 600 synthetic records, random row split, stratified, seed 42 |
How it works · three uses of language models
I use one pipeline three ways: extraction from messy text into structured records, a LoRA fine-tune that updates 1.1% of parameters, and semantic retrieval over past cases. The useful result is the negative one: the hybrid extractor's regex confidence gate chose the wrong records to escalate.
Notes · the corpus is synthetic and says so
- This ML demo sleeps after extended inactivity. First wake can take a moment; runs after that are quick.
- Every record in the corpus is synthetic and generated with a calibrated noise layer. The taxonomy and vocabulary use generic industrial maintenance concepts, not employer or customer records.
- The random row split can place shared generator templates in both train and test. The displayed metrics are within-generator results, not evidence of transfer to real CMMS text.
- The eight samples are mine: written for this page and labeled with what I intended. When the model disagrees with me, the page says so.
- Similarity search retrieves from the training corpus; the returned cases are synthetic records.
- Field extraction on this page runs the rule-based pass. The GPT-4o-mini extractor's 13% to 70% failure-mode comparison comes from a separate offline evaluation.