NeuroGrid Fault Risk Scoring Platform live

Medium-voltage fault risk inference with CI-minded training flow, artefact versioning, and live API serving.

End-to-End ML Model Serving CI Training Versioned Artefacts

View project

Forecast Studio live

Forecasting as a system: feature pipelines, backtesting harness, and deployment-ready outputs for planning workflows.

Time-Series Backtesting MLOps Monitoring-ready Outputs

View project

RAG Knowledge Copilot live

Retrieval tracing with source citations, guardrails enforcement, and automated evaluation harness for LLMOps workflows.

RAG LLMOps Evaluation Harness Guardrails

View project

Centrico LiveLab — End-to-End MLOps live

Complete MLOps pipeline: data ingestion → training → inference API → monitoring → CI quality gates.

Data Engineering MLOps Pipeline FastAPI Prometheus + Grafana CI Quality Gates

View project

Tabular Risk live

Interactive tabular risk scoring with client-side baseline model and optional API integration for deployed endpoints.

Client-Side ML Risk Scoring Transparent Baseline

View project

Visual Inspector live

Image analysis tool with local property inspection and optional API integration for deployed vision models.

Computer Vision Client-Side Analysis Image Processing

View project

Open Banking Transaction Enrichment live

AI-powered transaction categorization using NLP to extract merchant names, categories, and anomaly signals from raw Open Banking data.

NLP Categorization Merchant Extraction Multi-Language Anomaly Detection

View project

GLiNER Financial NER live

Zero-shot NER microservice for banking documents. Detects PII and financial identifiers using a DeBERTa encoder quantised to INT8 for CPU serving.

GLiNER ONNX INT8 FastAPI PII Detection Zero-shot NER

View project

EHD-Watch — Exocentric Homeostatic Agent live

A persistent agent whose primary welfare is world-directed. Monitors PM2.5 air quality in real time, deliberates on action selection, and exposes its full thought stream as it happens.

Exocentric Homeostasis BDI Deliberation Active Inference Groq / Llama 3.1 Cloudflare Functions

View project

EHD Paper — Can LLM Agents Care About the World?

Formal framework for persistent LLM agents with welfare anchored to an externally monitored world-state. Interventional hope term via do-calculus. Governance-transparent utility decomposition. Reproducible simulations.

arXiv 2026 Causal Inference LLM Agents Active Inference Governance

View evidence memo →