Safety Report Analysis Agent (Incident Precursor Detector)

Organizations collect thousands of free-text near-miss and safety observation reports. Buried within them are early warning patterns of serious incidents, but there is rarely time to read them all manually.

Develop an agent that reads free-text safety reports, classifies their risk level, extracts specific risk factors, and surfaces recurring patterns across reports.

1. Free-text report ingestion and parsing 2. Risk-level classification (low/medium/high) 3. Extraction of specific risk factors per report 4. Cross-report pattern detection (recurring themes, locations) 5. Dashboard summarizing trends for a safety officer

A report-parsing agent that extracts structured risk factors from free text A risk-scoring agent A pattern-detection layer across the full report set A summary dashboard

NLP/classification: LLM-based few-shot extraction and classification (no fine-tuning required) Analysis: Pandas for trend analysis Frontend: Streamlit / Plotly dashboard Data: Public OSHA/Kaggle near-miss datasets, or synthetic LLM-generated reports (clearly labeled as synthetic)NLP/classification: LLM-based few-shot extraction and classification (no fine-tuning required) Analysis: Pandas for trend analysis Frontend: Streamlit / Plotly dashboard Data: Public OSHA/Kaggle near-miss datasets, or synthetic LLM-generated reports (clearly labeled as synthetic)

Earlier identification of incident precursors before they escalate Reduced manual review workload for safety officers Actionable, location/department-level safety insights

Factory safety officers reviewing daily near-miss reports Construction site safety monitoring Lab safety compliance tracking

Quality of risk extraction and classification Usefulness of cross-report pattern detection Dashboard clarity