Applied ML Engineer — Sention

Energy / AI startup · Seed-backed (£1.2M)

Owned end-to-end applied ML systems for early fault detection, forecasting, and decision support in battery health — contributing directly to a £1.2M seed round.

Early Failure Detection (Unsupervised ML)

Problem: Battery failures were discovered late, increasing operational risk and costs.

• Built autoencoder-based anomaly detection to surface early degradation signals
• Owned the full ML pipeline: ingestion → training → inference → monitoring
• Deployed models via FastAPI + Docker, orchestrated with Airflow

→ 15% faster fault detection, enabling proactive maintenance and improving trust in ML-driven decisions

Failure Segmentation & Interpretability

Problem: Engineers lacked visibility into why batteries failed.

• Applied K-Means and DBSCAN clustering over sensor feature space
• Identified latent degradation patterns and created a failure-signature dataset
• Built visualizations to support engineering diagnosis

→ 20% improvement in interpretability and reduced time spent diagnosing failures

Battery Health Forecasting (Time-Series ML)

Problem: Manual grading slowed scaling and introduced inconsistency.

• Built LSTM and Transformer models for SoH / RUL prediction
• Modeled multi-sensor time-series data across battery lifecycles
• Achieved 7.8% MAPE in validation

→ 28% improvement in testing efficiency via automated grading pipelines

Internal AI Platform (Productized ML)

Problem: ML outputs were difficult for non-ML stakeholders to trust and use.

• Designed and shipped Sentinel, an internal AI platform
• Frontend: Next.js + Tailwind for investor- and operator-facing UI
• Backend: FastAPI + modular ML services
• Unified predictions, data, and business metrics in one system

→ 3× increase in internal ML adoption and faster iteration cycles

Business Impact

• Translated ML results into clear, investor-ready narratives
• Reduced faulty cell utilization by 12%
• Directly supported £1.2M seed funding