| --- |
| license: mit |
| language: |
| - en |
| library_name: transformers |
| pipeline_tag: text-classification |
| tags: |
| - email |
| - triage |
| - classification |
| - bert |
| - specific-ai |
| - gguf |
| base_model: google-bert/bert-base-uncased |
| --- |
| |
| # specific-AI/email-agent-triage |
|
|
| A compact **BERT** email triage classifier distilled with **[Specific AI](https://specific.ai)**. |
| It assigns each email to one of five action-oriented categories so agentic |
| workflows can decide whether to reply, archive, or take no action. |
|
|
| | | | |
| |---|---| |
| | **Task** | Single-label text classification | |
| | **Base model** | `bert-base-uncased` | |
| | **Training data** | ~15,000 examples | |
| | **License** | MIT | |
|
|
| ## Input format |
|
|
| Examples were trained on emails formatted as plain text with `From`, `Subject`, |
| and body (blank line between the headers and the body): |
|
|
| ```text |
| From: <from> |
| Subject: <subject> |
| |
| <body> |
| ``` |
|
|
| Pass inputs in this same shape at inference time for best results. |
|
|
| ## Labels |
|
|
| | Label | Meaning | Suggested next action | |
| |---|---|---| |
| | **URGENT** | Requires immediate attention (e.g. critical system failure, hard deadline right now). | Reply | |
| | **NEEDS_RESPONSE** | A task or response is owed, but it is not a drop-everything emergency. | Reply | |
| | **PROMOTIONAL** | Bulk mail, unsolicited promotions, or newsletters. | Archive | |
| | **PERSONAL** | Non-business, personal communications. | None | |
| | **FYI** | Informational only — the recipient should know, but no reply is required. | None | |
| |
| ## Evaluation |
| |
| Compared against **gpt-5.4-mini** as a teacher / baseline on the same evaluation set: |
| |
| | Metric | gpt-5.4-mini | SpecificAI | |
| |---|---:|---:| |
| | Accuracy | 0.693 | **0.720** | |
| | Precision | 0.810 | 0.763 | |
| | Recall | 0.693 | **0.720** | |
| | F1 score | 0.693 | **0.716** | |
| |
| ## Repository contents |
| |
| This card ships both a full Hugging Face checkpoint and GGUF-ready artifacts: |
| |
| - Full `BertForSequenceClassification` weights (`model.safetensors`) + tokenizer |
| - Head layers as NumPy files (`pooler_*.npy`, `classifier_*.npy`) for GGUF / Lemonade fusion |
| - Encoder GGUF: `bert-base-only.gguf` (CLS pooling; use with raw / unnormalized embeddings) |
| |
| ## Quick start — Transformers |
| |
| ```python |
| from transformers import AutoTokenizer, AutoModelForSequenceClassification |
| import torch |
| |
| model_id = "specific-AI/email-agent-triage" |
| tokenizer = AutoTokenizer.from_pretrained(model_id) |
| model = AutoModelForSequenceClassification.from_pretrained(model_id) |
| model.eval() |
| |
| text = """From: ops@example.com |
| Subject: Production outage |
| |
| Production is down — please escalate immediately.""" |
| inputs = tokenizer(text, return_tensors="pt", truncation=True) |
| with torch.no_grad(): |
| logits = model(**inputs).logits |
| pred = model.config.id2label[int(logits.argmax(-1))] |
| print(pred) |
| ``` |
| |
| ## Quick start — Lemonade + specific-ai-tools |
| |
| When running the GGUF encoder through Lemonade Server: |
| |
| ```bash |
| pip install specific-ai-tools |
| ``` |
| |
| ```python |
| from specific_ai_tools.embedding_heads import LemonadeEmbeddingClassifier |
| |
| classifier = LemonadeEmbeddingClassifier( |
| lemonade_model_name="user.email-agent-triage", |
| checkpoint="specific-AI/email-agent-triage:bert-base-only.gguf", |
| lemonade_base_url="http://localhost:13305", |
| ) |
| |
| text = """From: user@example.com |
| Subject: Billing question |
| |
| Please escalate this ticket to billing.""" |
| result = classifier.predict_one(text) |
| print(result.predicted_labels, result.predicted_confidences) |
| ``` |
| |
| See the Specific AI toolkit docs for llama-cpp and other embedding backends. |
| |
| ## Intended use |
| |
| - Email / inbox agent triage in production or on-device / CPU deployments |
| - Routing messages into reply / archive / no-action queues |
| |
| **Out of scope:** legal advice, medical triage, or safety-critical decisions without |
| human review. Labels reflect email workflow intent, not sender identity verification. |
| |
| ## About Us |
| |
| **[Specific AI](https://specific.ai)** is the automatic SLM distillation platform |
| that turns task prompts into production-grade small language models in days — |
| not weeks — so your subject matter experts can ship models without waiting on |
| scarce data-science bandwidth. |
| |
| We help enterprises move agentic AI from prototype to production with SLMs that |
| are typically **1,000×–10,000× smaller** than teacher LLMs, run in |
| **milliseconds** on CPUs or edge devices, and deliver the same or better |
| task quality at a fraction of the cost — self-hosted on your cloud or |
| downloaded for your own inference stack. |
| |
| **Prompt → Distill → Deploy.** Bring your prompt and data, drop them into |
| Specific AI, and get a validated small model ready to test and ship. |
| |
| Ready to create SLMs at scale? Visit **[specific.ai](https://specific.ai)**. |
| |
| ## License |
| |
| MIT — see [LICENSE](LICENSE). |
| |
| Copyright (C) 2026 Specific AI Inc. All rights reserved. |
| |
