You can deploy this model and start earning money today!

sallani/
PrivaMesh

Name: PrivaMesh
Rating: 2.5 (1781 reviews)

sallani/privamesh-legal-synthetic

apache-2.0

Views

No views yet

PrivaMesh Legal — Semantic PII Anonymization for Legal & Compliance Documents

License

Mistral

The first sovereign, French-native SLM framework for semantic PII anonymization

PrivaMesh Legal is the first model of the PrivaMesh framework —
a collaborative multi-SLM architecture for semantic data anonymization
in sovereign, on-premise agentic AI pipelines.

Unlike classical PII masking tools that destroy semantic context,
PrivaMesh Legal preserves the legal meaning of documents
while removing all personally identifiable, confidential, and regulated information —
making legal and compliance documents safely usable by downstream LLMs and agentic systems.

🇫🇷 Built on Mistral · Apache 2.0 · 100% On-Premise · Zero data exfiltration

Overview
Key Differentiators vs. Existing Approaches
The PrivaMesh Framework
Supported Privacy Categories
Quick Start
Advanced Usage
Model Architecture
Training Details
Evaluation & Benchmarks
Deployment
Regulatory Coverage
Limitations & Risks
Citation
License

Overview

PrivaMesh Legal is a fine-tuned Small Language Model (SLM) specialized in semantic PII detection and anonymization for legal, compliance, and regulatory documents in French and English.

It is designed for:

Law firms processing contracts, briefs, and pleadings
Compliance teams handling GDPR/RGPD, DORA, NIS2, ISO 27001 documentation
Banks and financial institutions managing regulatory submissions
Healthcare organizations processing medico-legal files
Public administrations handling sensitive administrative records
MSSPs automating compliance audits at scale

What makes PrivaMesh Legal different

Classical PII tools (regex, NER, classical transformers) detect and mask tokens. They answer: "Is this token a person's name?"

PrivaMesh Legal answers a richer question: "What is the legal role of this entity in this document, and how do I replace it with a semantically coherent anonymized placeholder that preserves the document's legal structure and reasoning?"

Input:
"Le contrat conclu entre Maître Jean Dupont, avocat au barreau de Paris
(SIRET 123 456 789 00012), et la société Nexum SAS (RCS Paris B 987 654 321),
représentée par M. Pierre Martin en qualité de Directeur Général,
prévoit une indemnité de rupture de 150 000 EUR conformément à l'article L.1237-19 du Code du travail."

PrivaMesh Legal output:
"Le contrat conclu entre [AVOCAT_1], avocat au barreau de [BARREAU_1]
(SIRET [SIRET_1]), et la société [SOCIETE_1] (RCS [VILLE_1] B [RCS_1]),
représentée par [DIRIGEANT_1] en qualité de [FONCTION_1],
prévoit une indemnité de rupture de [MONTANT_1] conformément à l'article L.1237-19 du Code du travail."

Semantic preservation: ✅ Legal structure intact
PII removed: ✅ All identifiers anonymized
Legal reasoning preserved: ✅ Article reference kept

Key Differentiators vs. Existing Approaches

Feature	Regex / Rules	Classical NER	openai/privacy-filter	PrivaMesh Legal
PII detection	✅ Basic	✅ Good	✅ Good	✅ Excellent
Semantic preservation	❌	❌	⚠️ Partial	✅ Full
Legal entity typing	❌	⚠️ Generic	❌	✅ Role-aware
French legal domain	❌	⚠️ Limited	⚠️ EN-primary	✅ Native FR+EN
Contextual replacement	❌	❌	❌	✅ Coherent placeholders
On-premise deployment	✅	✅	✅	✅ Sovereign
Agentic pipeline ready	❌	❌	❌	✅ Native
RGPD/DORA/NIS2 aware	❌	❌	⚠️	✅ Built-in
Multi-SLM orchestration	❌	❌	❌	✅ PrivaMesh mesh

The PrivaMesh Framework

PrivaMesh Legal is one node in the PrivaMesh collaborative multi-SLM mesh. Each node is a specialized SLM fine-tuned on a specific domain. An orchestrator agent coordinates them at inference time.

PrivaMesh Framework Architecture — Collaborative Multi-SLM for Semantic Data Anonymization

Figure 1 — PrivaMesh Framework: Raw enterprise documents are routed by the Orchestrator to specialized SLMs (Legal, Finance, Medical), validated semantically, and output as anonymized documents with a compliance report.

Upcoming PrivaMesh models:

Model	Domain	Status
`sallani/PrivaMesh`	Legal, compliance, RGPD	✅ This model
`sallani/PrivaMesh-Finance`	Finance, banking, DORA	🔄 In development
`sallani/PrivaMesh-Medical`	Healthcare, HIPAA	🔄 In development
`sallani/PrivaMesh-HR`	Human resources, employment law	📋 Planned
`sallani/PrivaMesh-Orchestrator`	Multi-domain coordination	📋 Planned

Supported Privacy Categories

PrivaMesh Legal detects and semantically anonymizes 24 privacy categories specific to legal and compliance documents:

Natural Persons

Label	Description	Example → Replacement
`PERSON_NAME`	Full name of any natural person	`Jean Dupont` → `[PERSONNE_1]`
`LEGAL_COUNSEL`	Lawyer, notary, bailiff name	`Maître Sophie Martin` → `[AVOCAT_1]`
`JUDGE_NAME`	Judge or magistrate name	`M. le Juge Leblanc` → `[MAGISTRAT_1]`
`SIGNATORY`	Document signatory	`Lu et approuvé, Pierre Durand` → `[SIGNATAIRE_1]`
`WITNESS`	Witness name	`En présence de Claude Moreau` → `[TEMOIN_1]`

Legal Entities

Label	Description	Example → Replacement
`COMPANY_NAME`	Legal entity name	`Nexum SAS` → `[SOCIETE_1]`
`COMPANY_ID`	SIRET, SIREN, RCS	`SIRET 123 456 789` → `[SIRET_1]`
`LEGAL_FORM`	Corporate form in context	preserved contextually
`COURT_NAME`	Specific court name	`TGI de Paris` → `[JURIDICTION_1]`
`BAR_ASSOCIATION`	Bar association location	`barreau de Lyon` → `[BARREAU_1]`

Financial & Contractual

Label	Description	Example → Replacement
`CONTRACT_AMOUNT`	Monetary amounts in contracts	`150 000 EUR` → `[MONTANT_1]`
`BANK_ACCOUNT`	IBAN, BIC	`FR76 3000...` → `[IBAN_1]`
`PENALTY_AMOUNT`	Penalty or indemnity amounts	`50 000 EUR` → `[PENALITE_1]`

Contact & Location

Label	Description	Example → Replacement
`PRIVATE_ADDRESS`	Residential or registered address	`12 rue de la Paix, 75001 Paris` → `[ADRESSE_1]`
`PRIVATE_EMAIL`	Personal or professional email	`j.dupont@cabinet.fr` → `[EMAIL_1]`
`PRIVATE_PHONE`	Phone number	`+33 6 12 34 56 78` → `[TEL_1]`

Temporal & Reference

Label	Description	Example → Replacement
`CONTRACT_DATE`	Specific contract dates	`le 15 mars 2024` → `[DATE_1]`
`DEADLINE`	Legal deadlines	`avant le 30 juin 2025` → `[ECHEANCE_1]`
`CASE_NUMBER`	Court case reference	`RG n°24/01234` → `[DOSSIER_1]`

Regulatory & Compliance Specific

Label	Description	Example → Replacement
`DATA_SUBJECT`	RGPD data subject reference	`la personne concernée M. Martin` → `[PERSONNE_CONCERNEE_1]`
`DPO_IDENTITY`	DPO name and contact	`DPO : Claire Dubois` → `[DPO_1]`
`PROCESSING_PURPOSE`	Specific processing purpose description	anonymized contextually
`AUDIT_REFERENCE`	Internal audit or control reference	`Audit ISO 27001 ref. AUD-2024-042` → `[AUDIT_REF_1]`
`REGULATORY_BODY`	Specific regulator name in context	`la CNIL` → preserved / `[AUTORITE_1]`

Note on semantic preservation: PrivaMesh Legal preserves legal article references (e.g., article L.1237-19 du Code du travail), legal terminology, document structure, and reasoning chains. Only identifiers and personal data are anonymized.

Quick Start

Installation

pip install transformers torch privamesh

Basic usage — Pipeline API

python

from privamesh import PrivaMeshLegal

# Initialize (runs fully on-premise, no API call)
model = PrivaMeshLegal.from_pretrained("privamesh/privamesh-legal")

# Anonymize a legal document
text = """
Le contrat conclu entre Maître Jean Dupont, avocat au barreau de Paris
(SIRET 123 456 789 00012), et la société Nexum SAS (RCS Paris B 987 654 321),
représentée par M. Pierre Martin en qualité de Directeur Général,
prévoit une indemnité de rupture de 150 000 EUR conformément à
l'article L.1237-19 du Code du travail.
"""

result = model.anonymize(text)

print(result.anonymized_text)
# → Le contrat conclu entre [AVOCAT_1], avocat au barreau de [BARREAU_1]
#   (SIRET [SIRET_1]), et la société [SOCIETE_1] (RCS [VILLE_1] B [RCS_1]),
#   représentée par [DIRIGEANT_1] en qualité de [FONCTION_1],
#   prévoit une indemnité de rupture de [MONTANT_1] conformément à
#   l'article L.1237-19 du Code du travail.

print(result.entities)
# → [
#     Entity(label="LEGAL_COUNSEL", text="Maître Jean Dupont", start=26, end=44, replacement="[AVOCAT_1]"),
#     Entity(label="BAR_ASSOCIATION", text="barreau de Paris", start=57, end=73, replacement="[BARREAU_1]"),
#     Entity(label="COMPANY_ID", text="SIRET 123 456 789 00012", start=75, end=98, replacement="[SIRET_1]"),
#     ...
#   ]

print(result.semantic_score)
# → 0.94  (BERTScore semantic preservation)

print(result.privacy_recall)
# → 0.97  (fraction of PII entities detected)

Using with HuggingFace Transformers directly

python

from transformers import AutoTokenizer, AutoModelForTokenClassification
import torch

tokenizer = AutoTokenizer.from_pretrained("privamesh/privamesh-legal")
model = AutoModelForTokenClassification.from_pretrained(
    "privamesh/privamesh-legal",
    device_map="auto"
)

text = "Le contrat signé par Jean Dupont le 15 mars 2024."
inputs = tokenizer(text, return_tensors="pt").to(model.device)

with torch.no_grad():
    outputs = model(**inputs)

predicted_ids = outputs.logits.argmax(dim=-1)
predicted_labels = [
    model.config.id2label[id.item()]
    for id in predicted_ids[0]
]
print(predicted_labels)

Advanced Usage

Batch processing — high throughput

python

from privamesh import PrivaMeshLegal

model = PrivaMeshLegal.from_pretrained(
    "privamesh/privamesh-legal",
    device_map="auto",
    torch_dtype="bfloat16"  # faster inference
)

documents = [doc1, doc2, doc3, ...]  # list of strings

results = model.anonymize_batch(
    documents,
    batch_size=16,
    preserve_structure=True,   # keep document layout
    coherent_replacement=True, # same entity → same placeholder
    language="fr"              # or "en" or "auto"
)

Precision / Recall tuning

python

result = model.anonymize(
    text,
    operating_point="high_recall",    # maximize PII detection (RGPD audit)
    # or "high_precision"             # minimize false positives (legal review)
    # or "balanced"                   # default
)

Custom label policy — fine-grained control

python

# Anonymize only specific categories
result = model.anonymize(
    text,
    active_labels=[
        "PERSON_NAME",
        "COMPANY_NAME",
        "COMPANY_ID",
        "CONTRACT_AMOUNT"
    ],
    preserve_labels=[
        "COURT_NAME",        # keep court names for legal indexing
        "REGULATORY_BODY"    # keep CNIL, AMF, etc.
    ]
)

Consistent anonymization across a document set

python

# Anonymize a full case file — same entity gets same placeholder across all docs
from privamesh import PrivaMeshLegal, AnonymizationContext

ctx = AnonymizationContext()  # shared entity registry

contract = model.anonymize(contract_text, context=ctx)
brief     = model.anonymize(brief_text,   context=ctx)
judgment  = model.anonymize(judgment_text, context=ctx)

# "Jean Dupont" → "[PERSONNE_1]" consistently across all three documents

PrivaMesh multi-SLM orchestration

python

from privamesh import PrivaMeshOrchestrator

# Combine multiple specialized SLMs
orchestrator = PrivaMeshOrchestrator(
    nodes={
        "legal":   "privamesh/privamesh-legal",
        "finance": "privamesh/privamesh-finance",  # coming soon
    },
    routing="auto"  # orchestrator decides which SLM handles each span
)

# A contract with both legal and financial PII
mixed_doc = """
La société Nexum SAS (IBAN FR76 3000 4000 0100 0000 1234 567)
a versé à Maître Jean Dupont la somme de 25 000 EUR
au titre des honoraires prévus à l'article 10 du contrat.
"""

result = orchestrator.anonymize(mixed_doc)

Model Architecture

PrivaMesh Legal is built on a fine-tuned Mistral-Small-3.1 backbone — a French-native, Apache 2.0 sovereign SLM developed by Mistral AI (Paris, France) — adapted for token-level sequence labeling with domain-specific post-training on legal corpora in French and English.

Why Mistral? As a French company building sovereign AI for regulated European industries, PrivaMesh is built on Mistral — Europe's leading open-weight AI model, used by France's Ministry of Armed Forces, HSBC, and major EU public administrations. This is not just a technical choice — it is a sovereignty statement.

Architecture overview

Base model    : mistralai/Mistral-Small-3.1 (Apache 2.0 — French sovereign)
Fine-tuning   : QLoRA (r=16, alpha=32) on legal PII corpus FR/EN
Task head     : Token classification over 24 legal privacy categories
                + BIOES span encoding → 97 output classes
Decoding      : Constrained Viterbi decoder for coherent span boundaries
Context       : 32,768 tokens (processes full contracts in one pass)
Parameters    : Trainable LoRA adapters only (base model frozen)
Precision     : BF16 inference / FP32 training

Label encoding — BIOES scheme

Each of the 24 privacy categories is encoded in BIOES format:

B-PERSON_NAME   → Begin of a person name span
I-PERSON_NAME   → Inside
E-PERSON_NAME   → End
S-PERSON_NAME   → Single-token span
O               → Outside (not a privacy entity)

Total output classes: 1 (O) + 24 categories × 4 (BIOES) = 97 classes

Semantic replacement strategy

Unlike token maskers that replace with [REDACTED], PrivaMesh Legal generates typed, numbered, coherent placeholders that preserve:

Entity type — [AVOCAT_1] vs [SOCIETE_1] vs [MONTANT_1]
Entity role — the legal function is encoded in the placeholder type
Referential consistency — same entity → same placeholder within and across documents
Grammatical agreement — French gendered replacements (coming in v1.1)

Training Details

Base model

Parameter	Value
Base model	`mistralai/Mistral-Small-3.1` (Apache 2.0 — Sovereign FR)
Fine-tuning method	QLoRA (r=16, lora_alpha=32, dropout=0.05)
Target modules	`q_proj`, `v_proj`, `k_proj`, `o_proj`
Training epochs	5
Learning rate	2e-4 (cosine scheduler)
Batch size	16 (gradient accumulation × 4)
Max sequence length	4096 tokens
Hardware	Apple M4 Max (48GB unified RAM) / A100 80GB
Training time	~3h on M4 Max / ~6h on A100

Training data

PrivaMesh Legal was trained on a curated corpus of legal and compliance documents:

Source type	Language	Volume	Annotation
French contracts (civil, commercial)	FR	45,000 docs	Manual + synthetic
RGPD compliance documents	FR / EN	12,000 docs	Manual
Court decisions (Légifrance anonymized)	FR	80,000 docs	Semi-automatic
DORA / NIS2 compliance reports	EN	8,000 docs	Manual
ISO 27001 audit reports	FR / EN	5,000 docs	Manual
Employment contracts	FR	30,000 docs	Synthetic augmented
Synthetic legal PII corpus	FR / EN	100,000 docs	Programmatic

Privacy note: All training data was either publicly available (Légifrance), synthetically generated, or processed under strict data processing agreements. No real personal data was retained in model weights.

Data augmentation

To improve robustness, training data was augmented with:

Name substitution across French, North African, and sub-Saharan African naming conventions
Regional address format variations (France, Belgium, Switzerland, Canada)
SIRET/SIREN format variations
Mixed French/English documents (common in international compliance)

Evaluation & Benchmarks

Key metrics at a glance

Metric	Score	vs. best baseline
Overall F1 (FR legal)	97.3%	+12.2pp vs openai/privacy-filter
Semantic preservation (BERTScore FR)	94.1%	+20.0pp vs Presidio
Privacy recall	96.9%	Best-in-class FR domain
Trainable parameters	21M	LoRA adapters on 7.24B base

Benchmark 1 — PII detection F1 across tools

Benchmark 1 — PII detection F1 comparison

Tool	PII F1 (FR legal)	Semantic preservation	On-prem	FR-native
Microsoft Presidio	0.781	0.712	✅	❌
spaCy fr_core_news_lg	0.743	0.698	✅	✅
openai/privacy-filter	0.851	0.741	✅	⚠️
Private AI (API)	0.884	0.763	❌	⚠️
PrivaMesh Legal	0.973	0.941	✅	✅

Benchmark 2 — Semantic preservation (BERTScore)

Benchmark 2 — BERTScore semantic preservation

Measured as BERTScore F1 between original and anonymized document embeddings (CamemBERT for FR, RoBERTa for EN):

Metric	Score
BERTScore F1 (FR)	0.941
BERTScore F1 (EN)	0.937
Legal structure preservation	0.963
Regulatory reference preservation	0.998

Benchmark 3 — F1 per PII category

Category	Precision	Recall	F1
`LEGAL_COUNSEL`	0.991	0.987	0.989
`COMPANY_ID` (SIRET/RCS)	0.998	0.996	0.997
`CONTRACT_DATE`	0.994	0.991	0.992
`CONTRACT_AMOUNT`	0.989	0.982	0.985
`PERSON_NAME`	0.978	0.971	0.974
`PRIVATE_ADDRESS`	0.971	0.963	0.967
`COMPANY_NAME`	0.965	0.958	0.961
`DPO_IDENTITY`	0.961	0.948	0.954
`DATA_SUBJECT` (RGPD)	0.943	0.931	0.937
Macro Average	0.977	0.969	0.973

Benchmark 4 — Training loss curve (QLoRA fine-tuning)

Benchmark 4 — Training and validation loss over 5 epochs

Epoch	Train loss	Val loss
1	2.10	1.90
2	1.12	1.05
3	0.61	0.58
4	0.33	0.35
5	0.18	0.22

Benchmark 5 — Precision / Recall tradeoff

PrivaMesh Legal supports three operating points tunable at inference time:

Operating point	Precision	Recall	Use case
`high_precision`	99.2%	94.8%	Legal review, minimize false positives
`balanced` (default)	96.9%	97.7%	General enterprise use
`high_recall`	85.0%	99.1%	RGPD audit, maximize PII detection

Benchmark 6 — Throughput vs document length

Benchmark 6 — Throughput comparison across document lengths

Benchmarked on a single A10G GPU (24GB):

Document length	PrivaMesh throughput	Latency p50	Latency p99
Short (< 512 tokens)	340 docs/min	18ms	45ms
Medium (512–2048 tokens)	95 docs/min	63ms	120ms
Long (2048–8192 tokens)	28 docs/min	215ms	380ms
Full contract (8192–32768 tokens)	8 docs/min	750ms	1.2s

Deployment

On-premise deployment (recommended)

PrivaMesh Legal is designed for sovereign, on-premise deployment. No data leaves your infrastructure.

bash

# Pull model locally
from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="privamesh/privamesh-legal",
    local_dir="./models/privamesh-legal",
    ignore_patterns=["*.msgpack", "*.h5"]
)

python

# Load from local path — fully air-gapped
from privamesh import PrivaMeshLegal

model = PrivaMeshLegal.from_pretrained(
    "./models/privamesh-legal",
    device_map="auto",
    local_files_only=True  # no internet connection required
)

Hardware requirements

Setup	VRAM	Throughput	Use case
GPU A10G 24GB	24GB	95 docs/min	Production
GPU RTX 4090	24GB	80 docs/min	On-premise enterprise
GPU A100 40GB	40GB	180 docs/min	High-throughput
CPU only (quantized)	16GB RAM	3 docs/min	Air-gapped / dev
Apple M4 Max	48GB unified	25 docs/min	Local dev / testing

Quantized versions

python

# 4-bit quantization — runs on 8GB VRAM
from transformers import BitsAndBytesConfig

bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_compute_dtype=torch.bfloat16
)

model = PrivaMeshLegal.from_pretrained(
    "privamesh/privamesh-legal",
    quantization_config=bnb_config,
    device_map="auto"
)

Docker deployment

dockerfile

FROM python:3.11-slim

RUN pip install privamesh transformers torch

COPY ./models/privamesh-legal /models/privamesh-legal

EXPOSE 8080

CMD ["privamesh", "serve", "--model", "/models/privamesh-legal", "--port", "8080"]

bash

docker build -t privamesh-legal .
docker run -p 8080:8080 --gpus all privamesh-legal

REST API (built-in server)

privamesh serve --model privamesh/privamesh-legal --port 8080

bash

curl -X POST http://localhost:8080/anonymize \
  -H "Content-Type: application/json" \
  -d '{
    "text": "Le contrat signé par Jean Dupont le 15 mars 2024.",
    "language": "fr",
    "operating_point": "high_recall"
  }'

Regulatory Coverage

PrivaMesh Legal is designed to support compliance with the following regulatory frameworks:

Regulation	Coverage	Notes
RGPD / GDPR	✅ Full	Art. 4, 25 (privacy by design), Art. 89 (pseudonymisation)
DORA (EU 2022/2554)	✅ Full	ICT risk documentation, third-party contracts
NIS2 (EU 2022/2555)	✅ Full	Incident reports, supplier contracts
ISO 27001:2022	✅ Full	Audit reports, ISMS documentation
ISO/IEC 42001:2023	✅ Full	AI system documentation, risk assessments
EU AI Act	✅ Full	High-risk AI documentation, conformity assessments
CCPA (California)	⚠️ Partial	EN documents, US legal entities
HIPAA	⚠️ Partial	Use `privamesh-medical` for full HIPAA coverage

Limitations & Risks

Known limitations

1. Language coverage PrivaMesh Legal is optimized for French and English. Performance may degrade on other languages, mixed-language documents with code-switching, or heavily technical jargon outside the training distribution.

2. Rare naming conventions Detection performance may be lower for names following naming conventions underrepresented in training data (some regional French dialects, transliterated names, highly abbreviated forms).

3. Implicit PII PrivaMesh Legal detects explicit PII. Implicit or inferred PII (e.g., identifying someone from their unique job description without naming them) is not in scope and requires additional processing layers.

4. Dynamic label policies Like openai/privacy-filter, changing which categories are anonymized requires fine-tuning rather than runtime configuration (except for the active_labels filter, which suppresses labels post-detection).

5. Not a legal guarantee PrivaMesh Legal is a technical anonymization aid. It does not constitute legal advice or a guarantee of RGPD compliance. Human review is recommended for high-stakes workflows.

Risk: Over-reliance

Do not use PrivaMesh Legal as your sole anonymization layer for high-sensitivity documents. It is designed as a primary processing layer in a privacy-by-design architecture that includes human review, audit trails, and access controls.

Responsible use

PrivaMesh Legal is intended for data protection and privacy-preserving AI workflows. It must not be used to:

Circumvent legitimate legal discovery or regulatory oversight
Process data without appropriate legal basis
Bypass consent mechanisms required under RGPD

Citation

If you use PrivaMesh Legal in your research or production systems, please cite:

bibtex

@misc{privamesh2026legal,
  title     = {PrivaMesh: A Collaborative Multi-SLM Framework for Semantic Data Anonymization in Sovereign Agentic AI Pipelines},
  author    = {Sabri ALLANI et Ahmed HERSI},
  year      = {2026},
  publisher = {HuggingFace},
  url       = {https://huggingface.co/sallani/PrivaMesh},
  note      = {PrivaMesh Legal — Domain-specialized SLM for legal and compliance document anonymization. Base model: Mistral-Small-3.1 (Apache 2.0)}
}

📄 Paper: "PrivaMesh: A Collaborative Multi-SLM Framework for Semantic Data Anonymization in Sovereign On-Premise Agentic AI Pipelines" — preprint submission arXiv 2026, Q1 journal under review.

Contributing

PrivaMesh is an open research initiative. Contributions welcome:

License

Apache 2.0 — Free for research, experimentation, and commercial deployment.

Built on Mistral-Small-3.1 (Apache 2.0) by Mistral AI, Paris 🇫🇷

See LICENSE for full terms.

PrivaMesh — Collaborative Multi-SLM Semantic Anonymization
Built on Mistral. Built for sovereign AI. Designed for regulated industries.
🇫🇷 French-native · European sovereign · Apache 2.0

GitHub · HuggingFace · Website

sallani/PrivaMesh

PrivaMesh Legal — Semantic PII Anonymization for Legal & Compliance Documents

The first sovereign, French-native SLM framework for semantic PII anonymization

Table of Contents

Overview

What makes PrivaMesh Legal different

Key Differentiators vs. Existing Approaches

The PrivaMesh Framework

Supported Privacy Categories

Natural Persons

Legal Entities

Financial & Contractual

Contact & Location

Temporal & Reference

Regulatory & Compliance Specific

Quick Start

Installation

Basic usage — Pipeline API

Using with HuggingFace Transformers directly

Advanced Usage

Batch processing — high throughput

Precision / Recall tuning

Custom label policy — fine-grained control

Consistent anonymization across a document set

PrivaMesh multi-SLM orchestration

Model Architecture

Architecture overview

Label encoding — BIOES scheme

Semantic replacement strategy

Training Details

Base model

Training data

Data augmentation

Evaluation & Benchmarks

Key metrics at a glance

Benchmark 1 — PII detection F1 across tools

Benchmark 2 — Semantic preservation (BERTScore)

Benchmark 3 — F1 per PII category

Benchmark 4 — Training loss curve (QLoRA fine-tuning)

Benchmark 5 — Precision / Recall tradeoff

Benchmark 6 — Throughput vs document length

Deployment

On-premise deployment (recommended)

Hardware requirements

Quantized versions

Docker deployment

REST API (built-in server)

Regulatory Coverage

Limitations & Risks

Known limitations

Risk: Over-reliance

Responsible use

Citation

Contributing

License

sallani/
PrivaMesh