Magic exist by Marta Loveguard - presentation.pptx
AI Driven Wi-Fi for the Bottom of the Pyramid
1. AI Driven Wi-Fi for the
Bottom of the Pyramid
BDNOG 17, 12 Dec – 15 Dec 2023
Mainul Hasan Alin, Founder, Janata Wi-Fi
2. Our Journey
• Built Wi-Fi Cloud that cost 1/10th of a big
vendor enterprise WIFI system.
• Innovated in distribution, partnered with ISPs,
DFS and Merchant networks.
• Innovated on Business Model, Advertisement
Based Free Wifi.
• Innovated on edge network automation
• Funded by APNIC Grant, Microsoft & Private
Funding
3. Public Wifi Experiments
• Mixture of Tea Stall, Mudi Dokaan & Bkash Shops
• Device is typically hosted within the shop
• Some of the nodes have very high concurrent usage
• Mobile App based device activation
CASE STUDY:
PERFORMANCE IN RAINKHOLA MIRPUR
• Operates 10 hours a day on average.
• Monthly 1000+ Ad views.
• Monthly 100+ Voucher Sold.
• Merchant gets USD .0025 per Ad view and USD .01 per
voucher.
Internet Connection
• 30-50 Meter Range
• 25 User Capacity
ISP
4. Operational Challanges
High Reliability Connections are Expensive.
Requires human interventions. Average issue solving
time is 10+ Hours.
Power failure is rampant with high cost of fuel.
Expensive Access Points.
Operation & Management for hundreds of Wifi Zones.
5. RETAIL CUSTOMER SERVICE CHALLENGES in BANGLADESH
Call
SMS
Digital
Others
• Billing, Package, Inquiry: Take
Note
• Fiber Movement, Bent/Cut:
Physical Support
• Majority, Bandwidth
Related: Remote and
Physical Support
5 minutes to pick up
3 Hours For Remote
Support to Call
10 Hours+ For Physical
Support to Arrive
Call Center Remote Support Physical Support
• Collects Data
• Debugs and Solves
• Or, forwards to Physical Support
• Or, non issue, convinces customer
• Challenges
• Users sometimes cant tell what
is ONU and what's its status.
• Remote access for command
line data collection requires a
bit sophisticated users. Its time
consuming.
30 Minutes For Remote
Support to Collect Data
• Fixes Fiber
• Restart Router, Some Config(Reset,
Change Admin/SSID user pass, wifi
channel, OS upgrade)
• Advice Customer on Changing from
Old Router
• Advice Customer on Placing the
Router from Right Place
• Challenges
• Convincing customer that it’s
a issue with old hardware
• Not many smart support
engineers that can debug
quick.
6. RETAIL CUSTOMER SERVICE AUTOMATION – Customer Interface
1 Minute to Navigate
2-5 Minute to Collect
Data
2-5 Minutes to
Resolve Issue
8. AI & ML Tools & Techniques
• Anomaly Detection
• For Time Series Data
• First Need to Establish a Baseline,
then look for anomalies
• Large Language Models
• Very Good at explaining
unstructured data.
• Works as a CoPilot to the Network
Support Engineer.
9. CXM System – Network Co-Pilot
Dashboard Co-Pilot
For Visual
Queues
As an Issue
Debugging
Assistant
10. Future Directions
• Data Collection from Non Conventional Agents.
• Work on Automated Solution Deployment and Confirmation
• Machine learning
• Automated issue detection and fixing
• Wifi Sensing
• Challenge: AI Talent is Expensive