AI Surgical Training

Project Overview

Surgical training programmes depend on reliable scheduling, but procedures vary widely in length and complexity. Trainees progress at different speeds, supervisors juggle competing demands, and theatre time is always limited. This project addresses that gap with a full-stack AI tool that brings predictive analytics directly into surgical training coordination.

Built as an MVP and presented to the Association of Surgeons in Training (ASiT), the platform estimates how long training tasks are likely to take, highlights sessions at risk of overrunning, and surfaces interpretable feedback to help coordinators plan more effectively — reducing wasted theatre time and improving trainee outcomes.

The Problem

When training sessions overrun, downstream trainees lose access to theatre time and supervisors face compressed teaching windows. Traditional scheduling relies on fixed time blocks that rarely reflect the variability of real procedures, skills assessments, or trainee experience levels. Coordinators often lack data-driven tools to anticipate delays before they happen.

Key Features

• Duration prediction: ML models estimate task completion time based on procedure type, trainee level, and historical session data.
• Overrun risk alerts: Flags sessions likely to exceed allocated slots so coordinators can adjust plans proactively.
• Interpretable feedback: Surfaces clear, actionable insights rather than opaque model outputs — critical in clinical training contexts.
• Training coordination dashboard: A React frontend gives supervisors a single view of upcoming sessions, risk levels, and scheduling recommendations.
• FastAPI backend: RESTful API serving predictions, session data, and model inference with low latency for real-time use.

Technologies Used

• Python & FastAPI for the prediction API and backend logic
• React for the interactive training coordination dashboard
• Scikit-learn / predictive analytics for duration and risk modelling
• Structured session data pipelines for training log ingestion
• CSS for responsive, clinical-friendly UI design

Results & Impact

The MVP was developed and demonstrated to ASiT, validating the concept with the surgical training community. The project showed that lightweight predictive analytics can meaningfully support scheduling decisions without requiring heavyweight infrastructure — a practical entry point for AI in clinical education workflows.

The work reinforced the importance of interpretability, domain-specific constraints, and building tools that fit existing clinical workflows rather than replacing them.

Links