Featured project

TransitLK
by Vaylen

From Smart Buses to Sovereign AI — a phased approach to nationwide mobility intelligence infrastructure. Converting Sri Lanka's 18,000+ buses into intelligent mobile nodes, then connecting them to adaptive traffic lights, smart highways and parking systems.

18,000+
Buses across Sri Lanka
SLTB ~5,000 · Private ~13,000
5M+
Daily bus passengers
~50% of all motorised trips
5
Platform phases
Buses → Traffic → Highways → Parking → Sovereign AI
2–3hrs
Lost daily per commuter
Due to unpredictable transport
Pilot Validated
Core AI models for Phase 1 have been tested on live SLTB buses — Bus NE-1810, Route 256/341-1 (Ratmalana Railway Station to Maharagama) from Moratuwa Depot. ETA prediction and multi-sensor crash detection have been validated in real Colombo conditions. Police and Fire Department have been briefed and provided positive feedback.

Sri Lanka's Transport Crisis

Sri Lanka's transportation ecosystem operates in information darkness. Over 18,000 buses move 5+ million passengers daily with zero real-time intelligence, no passenger communication, and fragmented safety systems — costing lives and billions in lost productivity.

A deeper issue is data sovereignty: foreign companies like Google and Uber extract Sri Lankan mobility data to overseas servers while local institutions have zero access to their own city's traffic intelligence — a strategic national vulnerability.

💸
Economic Drain
Commuters lose 2–3 hours daily to unpredictable transport. Fixed-timer traffic lights cause massive vehicle idling and fuel wastage costing the economy billions annually.
🚨
Safety Crisis
Bus accidents, fires, harassment and medical emergencies go undetected for critical minutes. Emergency services lack precise incident data and often arrive 10–20 minutes too late.
🌐
Data Sovereignty Gap
Foreign tech platforms own Sri Lanka's mobility data. Local institutions have no access to their own city's traffic intelligence — a strategic national vulnerability.
✈️
Tourist Experience
Visitors struggle with opaque public transport with no ETA, no English information, and unreadable route maps — defaulting to expensive private options.
Daily Commuter
No idea when the bus arrives or if it's coming at all.
Bus Operator
Fleet managed on gut feel — no driver behaviour data, no route efficiency.
Emergency Services
Accidents reported by bystanders with vague locations, 10–20 min after the event.
Traffic Police
Manual traffic direction during peak hours — fixed lights designed for 1990s volumes.
Tourist
No way to navigate buses unless a famous route. No ETA, minimal English info.
City Authority
No real-time data on congestion or movement — entire mobility dataset owned by foreign platforms.

A Five-Phase National Platform

TransitLK is a five-phase intelligent mobility platform that transforms fragmented transport infrastructure into a unified, AI-driven national network. The system operates across three levels: edge devices on buses and roads, AI/ML models in the cloud, and consumer and operator applications — creating a continuously learning urban intelligence layer.

01
🚌
TransitLK Smart Bus Network
Each bus becomes a smart node running three apps and an optional hardware kit
Pilot Complete
Real-time GPS tracking, AI crash detection, driver drowsiness monitoring, passenger safety alerts, ETA prediction, and crowd density estimation. Core AI models tested on live SLTB buses on real Colombo routes.
Passenger App (free)Driver AppAdmin App (SaaS)Smart Bus Hardware Kit2G SMS crash fallbackETA predictionCrash detectionDrowsiness detection
02
🚦
AI Traffic Light Network
Reinforcement learning replaces fixed-cycle traffic signals with adaptive, intelligent control
Models in Simulation
360° IR and 4K cameras with RL-based dynamic signal timing. CRNN licence plate recognition for automated violation detection. Emergency vehicle green corridors. Pedestrian and animal detection with IP68 ground LED strips.
RL signal optimisationCRNN plate recognitionEmergency corridorsPedestrian detectionYOLO vehicle detection20–40% congestion reduction
03
🛣️
Intelligent Highways Monitoring
ANPR-based contactless toll collection and automated speed enforcement on expressways
Models in Development
Entry/exit ANPR cameras for plate-based toll collection with direct bank deduction. Average speed calculation accounting for rest stops. Automated crash and breakdown detection with emergency dispatch. Target: Southern Expressway (Mahakumbura–Galle) pilot.
ANPR toll collectionAverage speed enforcementRest stop awarenessAutomated crash detectionRevenue share with RDA
04
🅿️
Smart Parking Infrastructure
Numberplate-based entry/exit with real-time occupancy pushed to the passenger app
Models in Development
Automated time and fee calculation via plate recognition. Real-time occupancy data pushed to the TransitLK passenger app so users can plan parking before arrival. Contactless payment and overstay violation detection.
Plate-based entry/exitReal-time occupancy APIContactless paymentOverstay detectionIntegrated with passenger app
05
🧠
Sovereign AI Mobility Infrastructure
Sri Lanka's first sovereign AI dataset — the foundation for autonomous transit and local LLM development
Future Scope
When Phases 1–4 are complete, the combined system becomes a 24/7 urban intelligence network. The data moat — years of real Sri Lankan mobility data — becomes the foundation for autonomous vehicle testing, an in-house Sri Lankan LLM trained on local context, and platform export to comparable developing nations.
City-wide security monitoringSovereign mobility datasetIn-house LLM trainingAutonomous vehicle testingInternational platform export

The Smart Bus System

📱
Passenger App
Free · Ad-supported
Destination-based route search with real-time ETA
Live bus crowding levels per route
Breakdown alerts with alternative route suggestions
Pickup requests for rare routes (within 30m of halt)
Lost item reporting and post-journey ratings
Sinhala, Tamil & English language support
🚌
Driver App
Free · Admin-activated
One-tap journey start/stop, logs GPS, accelerometer & gyroscope
Real-time incident alerts with colour-coded previews
Pickup request notifications within 300m
Critical crash alert — auto SMS via 2G to Police (119), Fire & nearest hospital
SMS includes GPS, crash type, passenger count, 60-second pre/post crash video link
Ad revenue share for highly rated drivers
🖥️
Admin App
SaaS · LKR 1,500–8,000/bus/mo
Real-time fleet dashboard with live positions
Trip analytics and incident log review
AI-powered route optimisation tools
Driver communication and performance tracking
Full 90-day footage archive (Enterprise tier)
Role-based access control (RBAC) — 4-tier
🔧 TransitLK Smart Bus Hardware Kit
Sold at ~50% manufacturing cost (loss leader) — recovered within 6 months via Pro/Enterprise subscription. Mandatory 6-month non-cancellation clause on all hardware agreements.
📷9× IR Cameras (interior monitoring)
⚙️18× Pressure Sensors (bumpers, sides, roof)
🚨Wireless Emergency Lightbars
🔋Ruggedized housing + backup battery
📶2G SMS fallback (offline crash alerting)
🛡️IP-rated for Sri Lankan conditions
Bus · Phase 1
ETA Prediction Engine
TensorFlow · XGBoost
Predicts bus arrival times at individual stops using GPS telemetry, historical patterns, and real-time traffic data. Validated on live Colombo routes.
Bus · Phase 1
Multi-Sensor Crash Detection
TensorFlow · OpenCV · XGBoost
Fuses accelerometer, gyroscope, and bumper/side/roof pressure sensor data. CV model provides secondary validation. Auto-dispatches emergency services within 30 seconds of impact.
Bus · Phase 1
Driver Drowsiness Detection
OpenCV · TensorFlow
Computer vision model monitoring driver facial behaviour via onboard IR cameras. Real-time alert to driver and admin dashboard when fatigue indicators detected.
Bus · Phase 1
Passenger Incident Detection
OpenCV · TensorFlow
Monitors interior camera feeds for falls, harassment, or medical emergencies. Demonstrated via TransitLK client product testing website with real-time model inference.
Traffic · Phase 2
Licence Plate Recognition
CRNN
Convolutional Recurrent Neural Network for automated plate recognition — supports violation detection (red light runners, speeding, hit-and-run) and toll collection.
Traffic · Phase 2
Adaptive Signal Control
YOLO · Reinforcement Learning
YOLO-based vehicle and pedestrian detection feeds a reinforcement learning model that dynamically adjusts signal timing — targeting 20–40% congestion reduction.

Full Technology Stack

Mobile & Web Apps
Flutter (Android/iOS/Web)
HTML / CSS / JavaScript
Kotlin
Backend & APIs
Python
Docker
REST APIs
Twilio (SMS)
AI/ML — Buses
TensorFlow
OpenCV
XGBoost
ETA prediction
Multi-sensor crash detection
Drowsiness detection
AI/ML — Traffic & Highway
CRNN (licence plate recognition)
YOLO (vehicle/pedestrian detection)
Reinforcement Learning (signal optimisation)
Synthetic Data
Blender (3D simulation)
Perspective warping scripts
AI-generated IR video
Edge Computing
Android/iOS devices (bus)
Mini PCs (traffic lights & toll gates)
IP68 embedded hardware
Cloud Infrastructure
Firebase
AWS / Azure
Local Sri Lankan data centres
Google Colab (GPU training)
CDN for passenger app
Simulation & Testing
SUMO (traffic simulation)
Hardware stress testing (Sri Lankan conditions)
OTA model update infrastructure
DevOps
GitHub
CI/CD pipelines
Agile Scrum sprints

How It Works In Practice

👤
Scenario A — Daily Commuter (Moratuwa → Colombo)
1
Opens TransitLK, enters destination. Sees three route options sorted by ETA and estimated fare.
2
Selects Route 100. App shows bus currently 2km away, ETA 8 minutes to their halt.
3
Bus breaks down mid-route. App immediately alerts: "Bus 100 broken down — next bus ETA 14 min. Route 101 available in 4 min."
4
Boards Route 101, rates the journey 4 stars. Driver receives rating and ad revenue share at month end.
🚨
Scenario B — Crash Response
1
Bus loses control and hits a tree. Multi-sensor fusion detects impact from accelerometer, gyroscope, and bumper pressure sensors. CV model provides secondary crash classification.
2
Driver app flashes red alert — driver must swipe within 15 seconds. If no acknowledgement, system auto-sends SMS via 2G to Police (119), Fire, and nearest hospital.
3
SMS contains: GPS coordinates, crash type, passenger count estimate, and a link to 60-second pre/post crash video.
4
Emergency services dispatched with precise data within 30 seconds of impact — vs the current 10–20 minute average for bystander reports.
🚦
Scenario C — AI Traffic Light (Phase 2)
1
RL model detects 47 vehicles queued on Galle Road northbound vs. 12 on the cross street.
2
System extends green phase by 22 seconds northbound, reducing queue by 34%. Simultaneously pre-clears a corridor for an ambulance detected 200m away.
3
A car runs a red light. CRNN captures the plate in 0.3 seconds. Violation logged with car model, colour, plate, and GPS timestamp — transmitted to Police enforcement system.

Six National-Level Outcomes

TransitLK isn't a startup app. It's a national infrastructure project targeting outcomes that matter to every Sri Lankan — from the daily commuter to the national government.

The ultimate ambition: when someone asks how to get from Colombo to Galle, they won't ask Google. They'll ask Vaylen AI — trained on millions of real Sri Lankan journeys, understanding local routes, schedules, safe areas, and cultural patterns of movement.

🏥
50% faster
Emergency response — automated crash detection cuts dispatch time
⏱️
30–60 min
Daily time saved per commuter via predictable arrivals & adaptive traffic
🌿
25% less CO₂
Adaptive signals reduce vehicle idling in pilot zones
🔒
100% local
All data processed and stored in Sri Lanka — no foreign platform access
🚀 Long-Term Vision (5–10 Years)
Year 5
50% of Sri Lankan urban commuters using TransitLK. 30% reduction in bus-related fatalities vs. pre-TransitLK baseline.
Year 7
Vaylen AI recognised as the authoritative source for Sri Lankan transportation queries — a sovereign alternative to Google for local mobility.
Year 10
TransitLK deployed in at least 3 other countries (Bangladesh, Nepal, Indonesia). Sri Lankan government using Vaylen data infrastructure for urban planning and policy decisions.

Built Together

🚌
Sri Lanka Transport Board (SLTB)
Project Partner · Data & Pilot Infrastructure
SLTB provides Vaylen with access to the national bus fleet, real operational data, and on-ground deployment infrastructure. The academic pilot was conducted on SLTB Bus NE-1810 operating Route 256/341-1 from Moratuwa Depot — validating core AI models on live, revenue-service buses.
💡
Creative Software
TransitLK Project Mentor · Technical Guidance
Creative Software mentors the TransitLK project specifically — providing engineering guidance, architecture review, and access to senior industry expertise. They are a leading Sri Lankan software company and their mentorship covers this project, not Vaylen as a whole organisation.

More Projects
Coming Soon.

TransitLK is just the beginning. As Vaylen grows, we're actively exploring new AI and ML initiatives beyond mobility — from data intelligence platforms to computer vision applications.

Have a project idea? We're always open to collaborations that push the boundaries of what's possible with AI in Sri Lanka and beyond.

Propose a Project → Our Services →
🚀
Next Project TBD
New AI initiatives are in early exploration.
Stay tuned