Digitize the logistic operations in Indonesia with the use of Advance Analytics
Reduce the wait time of freight at different touch points (entry gate, exit gate and docks) in a warehouse, in turn will help reduce load/unload throughput time
Enhanced visibility of logistic operations for management to identify opportunities for improvement and take actions accordingly
2. Agenda
1 Summary – Background and Understanding of requirements
2
3
Proposed Team Structure
Governance Mechanism
Proposed Solution
6
Project Plan
4
5 Support Required from Unilever
7 Commercials
3. SUMMARY OF REQUIREMENTS
B A C K G R O U N D & U N D E R S T A N D I N G
Automation Factory has been setup to provide intelligent automation services within Unilever landscape. AI CoE team is one of the portfolio
focused on solving business problems which cannot be handled by pure RPA. AI CoE team works on 3 major application areas; conversation,
vision and intelligence. Conversation caters to Chat Bot and NLP solutions. Intelligence caters to machine learning and deep learning solutions.
Vision caters to image and video analytic solutions.
In this presentation, we are proposing the end to end delivery of AI solution for Indonesia Truck Number Plate Project - business problem
assessment, conducting feasibility analysis, data gathering, data understanding, data preparation, model training & testing, model evaluation,
hosting and support to manage the risks/issues at each phase.
Background
Business Objective
1. Digitize the logistic operations in Indonesia with the use of Advance Analytics
2. Reduce the wait time of freight at different touch points (entry gate, exit gate and docks) in a warehouse, in turn will help reduce load/unload
throughput time
3. Enhanced visibility of logistic operations for management to identify opportunities for improvement and take actions accordingly
4. PROPOSED SOLUTION
Driver’s face is captured using
camera 1 at the entry gate
Number plate image is captured
using camera 2 at the entry gate
Face recognition to be done using Deep
Learning technique (CNN)
Image localization and OCR using
advanced deep learning methods
Truck picture is captured
using camera 3 at the docks
Truck Entry
Face is validated against the
driver master database.
Number plate is extracted and
matched against the database.
BP2A4904
Truck number is extracted and time
stamp is noted in the DMS
Checkpoint 1
Checkpoint 2
Truck at Docks
BP2A4904
On validation of face, truck
moves to checkpoint 2
Goes to manual
processing
Goes to manual
processing
Truck Exit
Truck picture is
captured using camera 4
at the docks
Goes to manual
processing
On validation of face & truck
number, time stamp is recorded in
TBS and truck moves to assigned dock
BP2A4904
Truck number is extracted and time
stamp is noted in the TBS
5. PROPOSED PROJECT PLAN (Continue..)
Program Plan & Governance
Communications, Stakeholder Impact Analysis and Change Management
Key
enablers
Functional Design : Jul 2019 – Sep 2019
Process re-design
(Functional)
Data hierarchy/
structure aligned
Business
Understanding
Stakeholder
alignment/ buy-in
Analysis &
Scoping of
the project
Technical
architecture
creation
based on the
solution
proposed
Future state Design:
Process & Technical
Sign off on
process & tech
design by
business
Sign off: Design &
Data
Analyze data
for quality &
bias
Hypercare & Support
Development
of model in
python
Project Lead
Identification
Develop
wireframes
Agreement on
data design
and availability
of system APIs
Data
Understanding
& Preparation
Hosting of
model on
Azure
Deployment
of model on
local
server/Edge
Integration of
model API with
UI & Systems
SCOPE IN THE PROJECT LIFE CYCLE
Data cleanup and
governance
process
Data owners and
access controls
Jan ‘19
Setup of Underlying Processes/SOPs
Provide support with issues
Integration
testing
OUR SCOPE
Development, Deployment & Integration: Oct 2019 – Jan 2020
Model Dev, Test
& Evaluation
Model Hosting
& Deployment
Integration &
Go-Live
Hypercare :Feb 2020 – Mar 2020
Re-training
model to
improve
accuracy
Go-Live in
Evaluate &
test the model
for accuracy
Development of User Interfaces
Hardware
requirements
and setup
Monitor the accuracy on a weekly
basis
Re-train the model with new data
6. PROPOSED PROJECT PLAN
• Servers, CPUs and cameras has
been procured & installed at the
site location
• List of software are installed in the
given system
Project Plan Oct-19 Nov-19 Dec-19 Jan-20 Feb-20
Phase Detailed activities 04-Oct 11-Oct 18-Oct 25-Oct 01-Nov 08-Nov 15-Nov 22-Nov 29-Nov 06-Dec 13-Dec 20-Dec 27-Dec 03-Jan 10-Jan 17-Jan 24-Jan 31-Jan 07-Feb 14-Feb 21-Feb
Face Detection
Dev & Hosting
Development & Unit Testing
Deployment & Hosting
Truck Plate Detection
Dev & Hosting
Development & Unit Testing
Deployment & Hosting
Integration & UAT
Stitching of models as per workflow
API Integrations with UI & Systems
Integration Testing
UAT
Hypercare Monitoring & Support (Retraining)
M1 M2 M3 M4
Pre-requisites • Full access to TSB & DMS
(APIs to be ready)
• User interfaces has been
developed
Deliverables
• Network connectivity is
established at the site location,
from server to camera with
stable network bandwidth
• Connection with cloud services
• Public IPs are available for the
servers
• Face Detection Model
(Accuracy > 90%)
• Deployment of model on
local server
• Model design document
• Truck Plate Detection
Model (Accuracy > 90%)
• Deployment of model on
local server
• Model design document
• Integration Test Scripts
• UAT Test Scripts
• Business Signoff on accuracy and results
• Monitoring Checklist
• Retraining Approach
Note
• Interaction between the following is established
• User interface and APIs
• User interface and camera
• 50 images has been provided for
each of the driver face
• Truck images has been captured
from past 1 month
• Truck License plate dataset must
be given which should be readable
without any dent
• License plate should not be broken
7. GOVERNANCE MECHANISM
Monthly
Review Meet
Frequency:
Monthly
Organizer:
Program Manager
Agenda:
Project updates,
results sharing
with Aadarsh
Weekly
Review Meet
Scrum meeting Weekly Status
Reporting
Frequency:
Weekly
Frequency:
Daily
Frequency:
Monthly
Organizer:
Program Manager
Organizer:
Business Analyst
Organizer:
Program Manager
Agenda:
Status of on-
going projects
and activities
with Kunal
Agenda:
Progress review,
support required
Agenda:
Market wise
project updates,
Achievements
Audience:
Program Lead,
Program Team
Audience:
Program Team,
Operating Team,
Technical
Delivery Team
Audience:
Operating Team,
Technical
Delivery Team
Audience:
Steering
Committee,
Program Team
*Proposed governance mechanism only
*Final structure to be discussed since some of these are already in place and run by the team currently
While some of
these forums are
already in place,
the team will
contribute to the
expectation of
the stakeholders
8. SUPPORT REQUIRED FROM UNILEVER
► Dedicated resources to drive the project from business point of view
► Seek buy-in of business users to a stronger participation in project
► Support in facilitating required data to proceed with the modelling in order to ensure timely delivery
Stakeholder
engagement
► Mechanism to cascade internally to users about the changes brought as part of program
► Support development of communication plan & elaborate key messages
Communication
► Ensure continuous support from infrastructure for smooth deployment of solutions
► Support timely resolution of issues/concerns
Infrastructure
► Regular feedback on performance
► Support timely resolution of issues/concerns
► Support in facilitating the right access to systems
Process
orchestration
Involvement Area Detailed responsibilities
9. PROPOSED TEAM STRUCTURE
PM/BA Data Scientist UI/UX/Analytics Developer
• Track deliverables, RAID and manage
governance
• Leadership reporting – Status, KPIs,
Challenges and Support Required
• Capture E2E business req. (PDD), gather
data, understand business requirement,
define problem and metrics and create
design docs
• Create delivery plan and conduct stand
ups and scrum meetings
• SPOC for all major communications (incl.
Go-Live and usage statistics etc.)
Key responsibilities • Data understanding and preparation
• Labelling the data
• Select and build model
• Train and test the model
• Deploy model
• Monitor performance and provide
necessary support
• Incorporate new data
• Retrain/Re-deploy
• Design and develop UI for user
management, dock dashboard, etc.
• Integrate the APIs of models and systems
wherever required
• Create PowerBI dashboard based on the
data collected
AI CoE Team
1 3 4
• Analyse the current technical/physical
architecture on site
• Create As-Is & To-Be architecture for the
deployment of hardware/software on the
site considering assumptions, risks and
possible failure scenarios
• Create solution architecture (SDD)
• Manage integrations and deployment
Technical Architect
2
10. COMMERCIALS
Project Cost: 62,000 Euro
Assumptions
1. The development will begin once we have signed off design document (by business)
2. All approvals (EA, Infosec, Legal, etc.) are in place
3. If any of the above approvals are not in place we need an email from project lead to pick up the process for development
4. Any deviation to the requirement documented in the design document will result in CR which will entail additional efforts and costs
5. Unilever will be charged based on the milestone achieved
6. If there are any delays in delivery due to reasons outside our scope such as below, we will immediately notify Unilever, put the delivery on hold and
charge for the efforts incurred till then:
i. Delays in UAT from business
ii. Unavailability of test data for internal testing/UAT
iii. Unavailability of infrastructure, public IPs, APIs, UI applications required for the process
iv. Infrastructure/application related issues
7. Clear success criteria to be agreed with Unilever and strictly adhered to
8. The project cost involves only implementation cost. Any cost pertaining to travel, boarding and lodging will be borne by the client if required for the
project.
Additional per day effort (for CRs/external delays): 175 Euro
Milestone based payment schedule:
• M1 – 30%
• M2 – 30%
• M3 – 30%
• M4 – 10%
11. OPERATING MODEL
Business understanding
Key
Activities
• Capture business
requirement
• Define problem
• Define metrics
• Develop solution
approach
• Gather data
Key
Performance
Indicators
Stage
Data Understanding Data Preparation Modelling Evaluation Hosting & Support
• Analyse data
• Access data quality
• Access data for bias
• Cleanse data
• De-Bias data
• Transform data
• Engineer features
• Select models
• Build models
• Train models with the
data prepared
• Test models for
accuracy
• Evaluate models and
select the best fit
model against the
problem
• Assess model
performance against
business objectives
(Internal Testing &
UAT)
• Host the model
• Monitor performance
• Incorporate new data
• Retrain/ Support
Key
Deliverables
• Process Design
Document/Approach
Note
• Effort Estimation and
Cost Analysis
• Business Sign off on
PDD/Approach Note
• Right data set to
proceed with data
preparation
• Transformed Data
• Feature Engineered
Data
• Data prep checklist
• Trained Model
• Model Retraining
Document
• Business Sign off on
model performance or
results (UAT Sign off)
• Go-NoGo Checklist
• Hosting architecture
• Monitoring checklist
• Retraining/Support
Approach Note
• PDD TAT
• Planned Vs Actual days
of PDD
• Model Accuracy
• No. of model retraining
• Model development TAT
• Planned Vs Actual days
of Model Development
• Hosting TAT
• Planned Vs Actual days
of Hosting
• PGLS TAT
• Planned Vs Actual days
of PGLS
• Data understanding and preparation TAT
• Planned Vs Actual days of Data understanding and
preparation
• UAT TAT
• Planned Vs Actual days
of UAT
• No. of issues in UAT
• No. of change requests
Business Analyst Data Scientist Data Scientist
Tech Architect Data Scientist
Business Analyst