1 of 32

AI-POWERED DIGITAL MARKETING OPTIMIZATION FOR SRI LANKAN SMEs (SINHALA & ENGLISH)

Date: Sep, 8, 2025

2 of 32

Our team

SLIIT

FACULTY OF COMPUTING

25-26J-301

1

Rajapakshe R S L

​

Jayasuriya J A D C G

​

Wijayarathna H.P.I.G.

Sandipa U.K.S

IT22082060

IT22112750

IT22116406

IT22095008

3 of 32

INTRODUCTION

Small and Medium Enterprises (SMEs) in Sri Lanka struggle with digital marketing due to a lack of expertise and resources. Current global solutions don't support local languages like Sinhala. Our project aims to bridge this gap by developing an AI-powered platform tailored for these SMEs.

​

This system will help them create content, automate customer engagement, and predict campaign performance in both Sinhala and English, empowering them to thrive in the digital economy.

SLIIT

FACULTY OF COMPUTING

25-26J-301

2

4 of 32

RESEARH QUESTIONS

SLIIT

FACULTY OF COMPUTING

25-26J-301

3

    • How can Artificial Intelligence be leveraged to create an accessible, multilingual (Sinhala & English) digital marketing optimization platform to empower Sri Lankan SMEs?
    • How can generative AI models be adapted to create context-aware marketing content in both Sinhala and English, tailored for the Sri Lankan market?
    • How can a conversational AI agent be designed to effectively manage customer interactions in Sinhala, English, and code-mixed language?
    • To what extent can machine learning models predict digital campaign success by analyzing multimodal data (images and text) from Sri Lankan SMEs?
    • What are the key features of a recommender system that can provide effective, personalized marketing strategies for diverse Sri Lankan SMEs?

5 of 32

INTRODUCTION TO THE PROBLEM

SLIIT

FACULTY OF COMPUTING

25-26J-301

4

    • SMEs in Sri Lanka are vital for economic growth but face challenges in digital marketing:
      • Limited expertise & resources
      • Poor access to intelligent tools
      • Language barriers (Sinhala not supported in most tools)

​

    • Results in:
      • Ineffective ad spending
      • Weak online branding
      • Delayed customer engagement
      • Missed opportunities in digital economy

​

6 of 32

SOCIAL OR RESEARCH IMPACT OF THE SOLUTION

SLIIT

FACULTY OF COMPUTING

25-26J-301

5

    • Social Impact:
      • Empowers local SMEs to compete digitally
      • Promotes inclusivity by supporting Sinhala + English
      • Reduces dependency on costly external marketing agencies
      • Enhances SME contribution to Sri Lanka’s digital economy

​

    • Research Impact:
      • Contributes to Sinhala NLP research (low-resource language)
      • Bridges AI + Marketing in a localized context
      • Opens opportunities for future multilingual AI solutions in Sri Lanka

​

7 of 32

PREVIOUS RESEARCH & SHORTCOMINGS, OUR APPROACH

SLIIT

FACULTY OF COMPUTING

25-26J-301

6

    • Previous Research:
      • Studies on digital marketing adoption by SMEs (Madurapperuma & Colombage, 2022)
      • Sinhala NLP surveys and models (De Silva, 2019; Senevirathne et al., 2020; Dhananjaya et al., 2022)
    • Shortcomings:
      • Not tailored for Sri Lankan SMEs’ real business needs
      • Lack of multilingual support (Sinhala + English, code-mixed)
      • No integrated platform (content, chatbots, predictions, strategy)
    • Our Approach:
      • Develop an AI-powered assistant with:
      • Multilingual content generator
      • Customer engagement chatbot
      • Campaign performance predictor
      • Personalized strategy recommender

​

8 of 32

CONSTRAINTS AND LIMITATIONS

SLIIT

FACULTY OF COMPUTING

25-26J-301

7

    • Data Limitations: Scarcity of large annotated Sinhala datasets
    • Technical Constraints: Need for fine-tuning models (GPT, BERT, ResNet) with local data
    • SME Constraints: Limited budgets, infrastructure, and digital literacy
    • System Limitations:
      • Focused mainly on social media marketing
      • Initial models may lack high accuracy due to limited training data

​

9 of 32

OVERALL SYSTEM DIAGRAM

SLIIT

FACULTY OF COMPUTING

25-26J-301

8

10 of 32

11 of 32

SLIIT

FACULTY OF COMPUTING

IT22095008

RAJAPAKSHE R.S.L

25-26J-301

PERSONALIZED MARKETING STRATEGY RECOMMENDER

Specialization: Information Technology

10

12 of 32

SLIIT

FACULTY OF COMPUTING

IT22095008

RAJAPAKSHE R.S.L

25-26J-301

12

PPERSONALIZED MARKETING STRATEGY RECOMMENDER

GAP & CREATIVE SOLUTION

13 of 32

SLIIT

FACULTY OF COMPUTING

IT22095008

RAJAPAKSHE R.S.L

25-26J-301

13

PERSONALIZED MARKETING STRATEGY RECOMMENDER

DOMAINS OF KNOWLEDGE

TECHNOLOGIES

DATA AVAILABILITY

​

ETHICAL CLEARANCE

    • Artificial Intelligence & Machine Learning – clustering SMEs by profile & performance.
    • Natural Language Processing (NLP) – Sinhala-English content understanding.
    • Digital Marketing & Data Analytics – strategy design, KPI evaluation.

​

    • Clustering - K-Means, DBSCAN.
    • Hybrid Recommender - Rule-based + ML models..

​

​

    • NLP - Hugging Face Transformers (Sinhala-English).
    • Development - FastAPI (backend), React.js (dashboard), Supabase (database)

​

    • Local SME surveys.
    • Social media pages & e-commerce sites (Daraz, Kapruka).

​

    • All SME data anonymized.
    • Secure storage & HTTPS/TLS encryption.
    • No personal identifiers stored.

​

    • Google Trends Sri Lanka.
    • Data Collect from Local Digital Marketing Agency

​

14 of 32

SLIIT

FACULTY OF COMPUTING

IT22095008

RAJAPAKSHE R.S.L

25-26J-301

14

PERSONALIZED MARKETING STRATEGY RECOMMENDER

EVALUATION & SUCCESS MEASUREMENT

​

15 of 32

SLIIT

FACULTY OF COMPUTING

IT22095008

RAJAPAKSHE R.S.L

25-26J-301

15

PERSONALIZED MARKETING STRATEGY RECOMMENDER

SYSTEM OVERVIEW

Real-World Scenario

    • Example: A fashion retailer enters profile.
    • System recommends:
      • Platforms: Instagram Reels (3/week), TikTok challenges, Facebook boosts.
      • Budget split: 70% Instagram ads, 20% TikTok, 10% Facebook.
      • Seasonal strategy: Avurudu promotion with giveaways.
    • SME runs campaign → results fed back → next recommendation more accurate.

16 of 32

SLIIT

FACULTY OF COMPUTING

IT22116406

Jayasuriya J A D C G

25-26J-301

INTELLIGENT CONTENT GENERATOR

Specialization: Information Technology

16

17 of 32

SLIIT

FACULTY OF COMPUTING

IT22116406

JAYASURIYA J A D C G

25-26J-301

17

INTELLIGENT CONTENT GENERATOR

GAP & CREATIVE SOLUTION

18 of 32

SLIIT

FACULTY OF COMPUTING

IT22116406

JAYASURIYA J A D C G

25-26J-301

18

INTELLIGENT CONTENT GENERATOR

DOMAINS OF KNOWLEDGE

​

    • Natural Language Processing (NLP): Using Large Language Models (LLMs) to understand and generate localized Sinhala-English text.
    • Machine Learning (ML): Fine-tuning models to learn and create culturally-aware content.

​

TECHNOLOGIES

​

    • Fine-tuned LLM (SinLlama): The first large language model specifically for Sinhala, adapted for Sri Lankan English context.
    • Subword Tokenization: Advanced technique to effectively process mixed-language text (Sinhala-English code-switching).

​

DATA AVAILABILITY

​

    • A custom "Sri Lankan Marketing Corpus" will be built from public social media data of successful local SMEs.

​

ETHICAL CLEARANCE

​

    • All collected data will be ethically scraped from public sources.
    • Data will be fully anonymized to remove personal identifiers before use.

​

EVALUATION PLAN

​

    • Quantitative: Using standard NLP metrics like BLEU/ROUGE scores for technical accuracy.
    • Qualitative: Conducting SME Focus Group testing for cultural relevance and user satisfaction.

​

19 of 32

SLIIT

FACULTY OF COMPUTING

IT22116406

JAYASURIYA J A D C G

25-26J-301

19

INTELLIGENT CONTENT GENERATOR

EVALUATION & SUCCESS MEASUREMENT

​

20 of 32

SLIIT

FACULTY OF COMPUTING

IT22116406

JAYASURIYA J A D C G

25-26J-301

20

INTELLIGENT CONTENT GENERATOR

SYSTEM OVERVIEW

    • The User: A saree shop owner in Kandy.

​

    • The Problem: Needs to create an engaging social media post for a weekly offer but isn't a marketing expert.

​

    • The Solution: She types a simple prompt into the Intelligent Content Generator:

"30% off sarees this week Kandy."

​

    • The Instant Results:

Localized English Post :

"Kandy fashion lovers… Enjoy 30% OFF on sarees this week. Hurry before stocks run out!"

Sinhala Post (Future Capability):

“මේ සතියේ හොඳම සාරි අඩුම මිලට… 🛍️ 30% ක විශේෂ වට්ටමක් අදම අත්විදින්න. ඉක්මන් කරන්න, අවස්ථාව සීමිතයි!”

​

21 of 32

SLIIT

FACULTY OF COMPUTING

IT22112750

WIJAYARATHNA H.P.I.G.

25-26J-301

CAMPAIGN PERFORMANCE PREDICOR

21

SPECIALIZATION: INFORMATION TECHNOLOGY

22 of 32

CAMPAIGN PERFORMANCE PREDICOR

GAP & CREATIVE SOLUTION

SLIIT

FACULTY OF COMPUTING

IT22112750

WIJAYARATHNA H.P.I.G.

25-26J-301

22

23 of 32

23

CAMPAIGN PERFORMANCE PREDICOR

KEY DOMAINS

    • Artificial Intelligence & Machine Learning
    • Natural Language Processing (NLP) – Sinhala-English content understanding.
    • Digital Marketing & Data Analytics
    • Explainable AI (XAI)

​

​

TECHNOLOGIES

    • Transformer models (BERT / multilingual)
    • Python (TensorFlow / PyTorch)
    • Sinhala + English dataset creation
    • API integration for deployment
    • Explainability – SHAP, LIME for feature importance and interpretation.

​

DATA AVAILABILITY

    • Local SME surveys.
    • Social media pages & e-commerce sites (Daraz, Kapruka).
    • Google Trends Sri Lanka.
    • Synthetic + transfer learning from global datasets.

​

ETHICAL CLEARANCE

    • All SME and campaign data will be anonymized before use.
    • Data will be stored in secure databases with encryption (HTTPS/TLS).
    • No personal identifiers stored.

​

SLIIT

FACULTY OF COMPUTING

IT22112750

WIJAYARATHNA H.P.I.G.

25-26J-301

24 of 32

CAMPAIGN PERFORMANCE PREDICOR

SLIIT

FACULTY OF COMPUTING

IT22112750

WIJAYARATHNA H.P.I.G.

25-26J-301

24

EVALUATION & SUCCESS MEASUREMENT

​

25 of 32

CAMPAIGN PERFORMANCE PREDICOR

SYSTEM OVERVIEW DIAGRAM

SLIIT

FACULTY OF COMPUTING

IT22112750

WIJAYARATHNA H.P.I.G.

25-26J-301

25

Real-World Scenario

    • Example: A small bakery in Colombo wants to promote a new cake design.
    • They upload the a Sinhala or English content, and select Facebook as the platform with a posting time.
    • The Campaign Performance Predictor analyzes the content and shows:
    • Expected reach: 2,000 people
    • Likes: 300
    • Comments: 44
    • Shares: 23
    • Quality score: 78/100 (Good)
    • The system also explains:
      • Caption is positive but too short → add hashtags.
      • Posting time could be improved → evening 7–9 pm gives higher engagement.

​

26 of 32

SLIIT

FACULTY OF COMPUTING

IT22082060

SANDIPA U.K.S

25-26J-301

CUSTOMER ENGAGEMENT

CHATBOT

SPECIALIZATION: INFORMATION TECHNOLOGY

26

27 of 32

SLIIT

FACULTY OF COMPUTING

IT22082060

SANDIPA U.K.S

25-26J-301

27

CUSTOMER ENGAGEMENT CHATBOT

GAP & CREATIVE SOLUTION

28 of 32

SLIIT

FACULTY OF COMPUTING

IT22082060

SANDIPA U.K.S

25-26J-301

28

CUSTOMER ENGAGEMENT CHATBOT

KEY DOMAINS

    • Natural Language Processing (chat understanding)
    • Machine Learning (training chatbot)
    • Human–Computer Interaction
    • Web Integration (connect with website/ Messanger)

​

TECHNOLOGIES

    • Transformer models (BERT / multilingual)
    • Python (TensorFlow )
    • Sinhala + English dataset creation
    • API integration for deployment

​

DATA AVAILABILITY

    • Local SME surveys.
    • Social media pages & e-commerce sites (Daraz, Kapruka).
    • Google Trends Sri Lanka.
    • Synthetic + transfer learning from global datasets.

​

ETHICAL CLEARANCE

    • Anonymized customer data only
    • No sensitive/private info used
    • Privacy and confidentiality followed

​

29 of 32

SLIIT

FACULTY OF COMPUTING

IT22082060

SANDIPA U.K.S

25-26J-301

29

CUSTOMER ENGAGEMENT CHATBOT

EVALUATION & SUCCESS MEASUREMENT

30 of 32

    • Customers visit the business website or social media and type a question or request in bot.
    • Chatbot receives the message.
    • NLP engine understands the intent in Sinhala or English.
    • Chatbot retrieves relevant information from the business database (orders, product info, FAQs).
    • Chatbot sends a clear, fast reply to the customer on the same platform.
    • Example: “Your order #123 will arrive tomorrow” / “ඔබගේ ඇනවුම හෙට ළඟා වේ.”

​

SLIIT

FACULTY OF COMPUTING

IT22082060

SANDIPA U.K.S

25-26J-301

30

CUSTOMER ENGAGEMENT CHATBOT

SYSTEM OVERVIEW DIAGRAM

Real-World Scenario

31 of 32

SLIIT

FACULTY OF COMPUTING

25-26J-301

31

AI-POWERED DIGITAL MARKETING OPTIMIZATION FOR SRI LANKAN SMES (SINHALA & ENGLISH)

PROJECT COMMERCIALIZATION & BUSINESS MODEL

ABILITY OF COMMERCIALIZATION

INVESTMENT / COST & RECOVERY

TARGET MARKET & CUSTOMER PROFILE

VALUE PROPOSITION

INTELLECTUAL PROPERTY RIGHTS (IPR)

    • SaaS-based solution for SMEs in Sri Lanka & South Asia
    • Affordable subscription → Free tier + Premium features

​

    • Labor: Development team (4 members, 1-year project)
    • Material: Cloud hosting, APIs, data collection tools (~LKR 500,000)

​

    • SMEs (retail, fashion, food, education, services)
    • Owners with limited marketing expertise
    • Affordable & bilingual (Sinhala + English)
    • End-to-end personalized strategies
    • Feedback-driven improvements
    • Easy-to-use dashboard
    • Proprietary recommendation engine
    • Trademark & SaaS branding
    • Cost Recovery (Pricing Model):
      • Free basic plan (trial adoption)
      • Premium subscription ~ LKR 2,000–3,000/month
      • Enterprise custom plans for larger SMEs

​

    • High demand: >1 million SMEs in Sri Lanka with limited access to AI-driven marketing

32 of 32

THANK YOU