Contact person: Business Client

Phone:Show

Email:Show

Location: Stockport, United Kingdom

Budget: Recommended by industry experts

Time to start: As soon as possible

Project description:
"I’m building T-LAIOR, an orchestration layer that governs a network of agentic and model-based systems. As part of this, I need a Vertex AI–native machine learning component that can continuously monitor, classify and enforce regulatory and security compliance across multiple autonomous agents.

The incoming data stream contains security-governance, telemetry and compliance events. Some fields are cleanly structured; others are semi-structured or free-text alerts generated by agents and sub-systems. Your job is to build a model and supporting pipeline that can handle mixed-format data, detect potential policy or compliance deviations, and produce a fully auditable trust signal that can be consumed by the T-LAIOR trust layer.

Core Requirements

The full workflow—data ingestion, preparation, feature extraction, model training, tuning, evaluation and deployment—must run entirely inside Vertex AI using managed datasets, pipelines and model registry. This is essential because the resulting component will become an SLM-level governance module inside T-LAIOR.

The model’s primary goals are to:

Identify probable compliance breaches in real time

Provide clear, traceable explanations for both compliant and non-compliant classifications

Integrate cleanly with the T-LAIOR Trust Layer, which requires explainability metadata and consistent output schemas

What You Need to Bring

You should have strong experience with:

Vertex AI Pipelines, Workbench, AutoML or custom model training

BigQuery + Cloud Storage ingestion patterns

Mixed-format feature engineering (structured + unstructured text)

Vertex Explainable AI and model interpretability techniques

Security-first GCP engineering (IAM, VPC-SC, CMEK, service perimeter design)

Experience with compliance analytics, cyber governance, policy modelling or risk scoring is a major plus.

Deliverables

1. A reproducible Vertex AI Pipeline (Python SDK) that:

Ingests data from BigQuery + Cloud Storage

Cleans, validates and featurises mixed-format records

Trains and tunes the compliance classification model

Logs metadata to Vertex AI and produces lineage for the trust layer

2. A trained model registered in Vertex AI Model Registry

Ready for batch and online prediction

Emits explanations compatible with the T-LAIOR trust schema

3. A full evaluation package including:

Precision, recall, ROC-AUC on a held-out test set

Explainability outputs (Vertex Explainable AI)

Bias, drift and consistency checks if applicable

4. A production-ready deployment plan:

Endpoint configuration with IAM, VPC-SC, CMEK

Online and batch inference setup

Monitoring and alerting configuration

5. Documentation packages:

Step-by-step setup instructions to rerun in a clean GCP project

Integration notes for plugging the model into the T-LAIOR governance workflow

If you’ve done similar work in regulated environments, ML for security/compliance, or agent governance inside Vertex AI, highlight it—this expertise will be particularly relevant." (client-provided description)