Shariful Islam Foysal

Architected and implemented an enterprise-grade conversational AI solution that transforms natural language questions into SQL queries, enabling non-technical users to access complex database insights. Built using LangChain and Amazon Bedrock, the solution implements Retrieval-Augmented Generation (RAG) architecture to ensure accurate query generation based on business context and data schemas.

The system features a user-friendly Slack bot interface for query requests and a React web application for knowledge base management. Implemented robust security controls through authentication and built automated guardrails for query validation and blocking harmful content. The solution is highly scalable and deployed in the cloud.

Achieved significant business impact by reducing query development time by 2,500 hours monthly and democratizing data access across the organization. The system serves 500+ users and maintains high accuracy through continuous feedback mechanisms and automated performance monitoring.

Designed and implemented an automated dynamic pricing system that optimizes revenue and market efficiency for a ride-hailing platform. The solution combines real-time supply-demand forecasting with intelligent price adjustment algorithms to maximize driver availability and rider satisfaction.

Built a robust machine learning pipeline using time-series forecasting models to predict hyperlocal demand and supply patterns. Implemented an adaptive pricing algorithm considering multiple factors, including historical patterns, current market conditions, weather impact, and special events. The system processes real-time data streams to adjust prices dynamically, ensuring optimal market balance across different city zones.

The architecture leverages containerized microservices for scalability, with automated model retraining pipelines and A/B testing capabilities. Implemented comprehensive monitoring and alerting systems to track key performance metrics,s including driver utilization, rider wait times, and market equilibrium indicators.

Architected and developed a machine learning (ML)-based demand forecasting system for new product launches in retail. The solution utilizes LightGBM to predict 6-month aggregated demand for products with long lead times (6-12 months), enabling data-driven inventory and sourcing decisions.

Built an end-to-end ML pipeline that processes historical sales data, product attributes, and market indicators to generate accurate demand predictions. The system incorporates multiple granularity levels of product attributes and seasonality patterns to handle the cold-start problem inherent in new product forecasting. Implemented a specialized version (2.0) for newly launched products, with modified feature engineering and hyperparameter optimization.

The solution significantly improved forecast accuracy over traditional heuristic methods, enabling better inventory planning and reducing stockouts while optimizing working capital. The system processes forecasts across multiple European marketplaces, supporting strategic product design and supply chain planning decisions.