Safer bus journeys through explainable, multi-modal AI
CITOS watches for fatigue, distraction, aggressive driving and traffic violations in real time — and explains the law behind every alert through a GraphRAG assistant that speaks Sinhala, Singlish and English.
- Research components
- 6
- Languages supported
- 3
- Hallucination-control layers
- 5
- Distraction classes
- 10
Six components, one safety picture
Each component tackles a different cause of road accidents. Together they give fleet operators a complete, explainable view of every driver and every trip.
Verifiable GraphRAG
A question-answering assistant that lets drivers, fleet managers and officers ask about traffic law and driver records in Sinhala, Singlish or English — and only answers with claims it can trace back to a knowledge graph.
Fatigue detection
Real-time drowsiness detection from a cabin camera, running face-landmark analysis directly in the browser so video never needs to leave the vehicle.
Distraction detection
A deep-learning classifier that recognises distracting activities from cabin imagery and pairs them with the road scene to judge how risky the distraction is at that moment.
Behaviour analytics
IMU-based telemetry that classifies driving style in real time, estimates passenger comfort and tracks component wear for predictive maintenance.
Violation detection
Dash-camera and GPS analysis that detects traffic-light and pedestrian-crossing violations and judges speed against the real context — school zones, rain and route geometry.
Risk scoring
A fair driver score that replaces flat point deductions with an exposure-normalised, Bayesian-smoothed and time-decayed risk rate.
Diagnose. Verify. Personalize.
Our multilingual traffic-law assistant never answers from memory alone. Every claim is traced to a knowledge graph of drivers, violations and the Motor Traffic Act — and when the evidence isn't there, it says so.
- 01
Ask
Voice or text in Sinhala, Singlish or English. Speech is transcribed first.
- 02
Guard
Prompt-injection, PII, toxicity, relevance and Cypher-injection guardrails screen the input.
- 03
Understand
A LoRA-tuned Qwen2.5 model normalises Singlish, and XLINK binds names to graph values.
- 04
Retrieve
Text-to-Cypher and bridge retrieval pull operational and legal evidence from Neo4j.
- 05
Diagnose
The evidence-sufficiency gate abstains before generation if the rows can't answer the question.
- 06
Generate
Gemini writes a grounded answer, personalised with the caller's own graph profile.
- 07
Verify
GAFL, Chain-of-Verification and conformal abstention check every claim before it is shown.
From raw signals to grounded answers
A layered architecture keeps sensing, analytics and reasoning independent, so each research component can evolve on its own.
Sense
Cabin and dash cameras, an IMU motion sensor and GPS capture the driver, the vehicle and the road.
Analyse
Vision models, Random Forests and YOLO detectors turn raw signals into fatigue, distraction, behaviour and violation events.
Reason
Events flow into a Neo4j knowledge graph linked to traffic law, where a verifiable GraphRAG assistant answers questions.
Act
Drivers get real-time alerts, managers get fair risk scores and everyone gets grounded answers in their own language.
Interested in collaborating or piloting the system?
We'd love to hear from fleet operators, researchers and road-safety authorities.