Helping commerce chatbots understand routine customer questions.
A Flask service designed to help marketplace and e-commerce chatbots interpret common questions and automate parts of routine customer-support replies using language rules, curated datasets and lightweight statistical models.
The problem
Marketplace and e-commerce clients received recurring questions about products, specifications, orders, inventory, locations and contact information. The goal was not to replace customer-support teams, but to give chatbots enough structured understanding to automate parts of common replies and support the wider conversation workflow. This work preceded my formal AI/ML research, so I approached it as a practical engineering and data problem.
What I was responsible for
- 01Reviewed real user queries and organized examples into commerce-focused intent, entity and sentiment datasets.
- 02Defined intents for common support needs, including product questions, purchases, specifications, order status, inventory, locations and organization contacts.
- 03Built Flask endpoints for intent and entity detection, sentiment analysis, date/time extraction and email parsing so the surrounding chatbot could consume structured results.
- 04Combined tokenization, language and part-of-speech processing, dictionary and conditional rules, and TF-IDF-based statistical classification.
- 05Supported English, Bengali and Banglish inputs and stored user context in MySQL as query patterns evolved.
What happened
The preserved repository contains separate Flask resources for NLU, sentiment, email and date/time parsing; commerce-focused intent categories; English, Bengali and Banglish language resources; curated product, color and location data; TF-IDF model artifacts; and MySQL-backed user context. It documents a working support component within a broader chatbot workflow rather than a fully autonomous customer-service system.