AI-Powered Multi-Vendor E-Commerce System
Thesis project (In Development): ML for personalized recommendations, dynamic pricing, AI-driven customer assistance and detection of scams, spam, malicious content and vulnerabilities.
Thesis
Feni University
3
AI pillars
In Dev
status
The brief.
-
problem
Marketplaces struggle with relevance (recommendations), pricing competitiveness and trust (scams/spam/malicious content). Academic goal is to combine software engineering + cybersecurity + AI in one evaluatable system.
-
overview
Developing ML models for personalized product recommendations, dynamic pricing strategies and AI-driven customer assistance within a multi-vendor e-commerce shell. Exploring detection of scams, spam, malicious content and security vulnerabilities using machine learning.
-
solution
Building a Laravel commerce base and integrating Python ML services: recommendation engine, dynamic pricing model, AI customer assistant and classifiers for scams/spam/malicious content. Investigating model training, evaluation, deployment and application integration.
-
role note
Academic thesis project — investigation, model training/evaluation, deployment and end-to-end platform integration.
-
business goal
Design an architecture for integrating AI-based detection into web platforms while combining web tech, DB management and AI models into a scalable end-to-end solution.
Modules I worked on
Stack
Laravel Python AI MySQL TensorFlow OpenAICase study.
Thesis Project — In Development · Feni University · Combining software engineering, cybersecurity and AI.
Project Overview
Developing an end-to-end multi-vendor e-commerce system that integrates AI models directly into the commerce architecture. The platform combines Laravel web technologies, database management and Python-based ML services.
AI Features
- Personalized product recommendations — collaborative + content-based ranking
- Dynamic pricing strategies — data-driven pricing adjustments
- AI-driven customer assistance — contextual shopping help
- Scam / spam / malicious content detection — ML classifiers for trust & safety
- Security vulnerability detection — ML-assisted code/content scanning
Engineering Focus
Investigating model training, evaluation, deployment and application integration. Designing an architecture for integrating AI-based detection into web platforms so that software engineering, database management and AI models scale together.
Status
In Development — thesis defense scheduled 2026. The system is being built as a scalable, evaluatable whole rather than a bolted-on demo.
Implementation notes
The architecture keeps Laravel as the commerce core and moves model inference into separate Python services, so training and deployment cycles stay independent from feature releases. The recommendation pipeline starts content-based (embeddings over product attributes) and blends in collaborative signals as interaction data accrues, which is what keeps recommendations useful during the cold-start period. The dynamic pricing module scores price elasticity from historical orders and proposes adjustments inside vendor-defined floors and ceilings, never overriding them.
Trust and safety work is classifier-first: scam and spam detection run as ensemble classifiers over listing text and user behaviour, with vulnerability scanning handled as a separate pipeline over code and content payloads. Every AI feature ships with an evaluation harness so prompt or model changes are gated on measurable precision before they reach production, and all inference calls pass through a queue with cost metering so experiments cannot silently burn budget.