Industry AI Platforms
TRIDENT — Retail Intelligence & Decision Support Platform
Architected an AI-powered retail intelligence platform combining dual-tower deep learning, ensemble learning, and graph intelligence to transform large-scale FMCG transaction data into actionable insights — supporting assortment optimization across 2.5M+ retail outlets.
Tech: Deep Learning, Graph Intelligence, Recommendation Systems, LLM-Assisted Feature Engineering
Q-Commerce Intelligence Engine
Built scalable recommendation models for quick-commerce assortment optimization, combining interaction-based embeddings with gradient-boosted models to predict outlet-level product fit — deployed across multiple FMCG categories including Oral Care, Hair Care, Personal Wash, and Skin Care.
Tech: LightGBM, Matrix Factorization, Feature Engineering, Cloud Data Pipelines
Big Bets — Twin Training for Prioritized Decisions
Co-developed Twin Training, a model-agnostic framework that applies a global learning phase followed by a priority-focused refinement phase, surfacing gains on high-impact minority cases without degrading overall model calibration. Published at FLAIRS-39 (2026).
Tech: Gradient-Boosted Trees, Neural Networks, Explainable AI
LLM-Assisted Feature Engineering
Developed a framework using Large Language Models to automate discovery of business-relevant features from raw data, reducing manual feature engineering effort while improving model explainability. Presented as an invited tutorial at ACM DocEng 2025, University of Nottingham, UK.
Tech: LLMs, Automated Feature Discovery, Document Understanding
Digital CEO — Voice-Enabled Executive Decision Intelligence
A voice-enabled assistant that lets leaders ask questions and get curated insights across business systems instantly, using embedding models, Vector DBs, LLMs, and RAG — turning natural-language queries into fast, data-driven decisions.
Tech: LLMs, RAG, Vector Databases, Embeddings
AI Tutor — Personalized Learning Companion
A real-time learning assistant for educational video content, giving students and educators personalized feedback that turns passive viewing into an interactive learning experience.
Tech: GenAI, Personalization, Real-Time NLP
Product Knowledge Graphs
Built knowledge graphs from customer reviews and conversational data using NLP and graph techniques, surfacing product preferences, sentiment, and market trends that directly informed business decisions and increased profitability.
Tech: Knowledge Graphs, NLP
Applied Research Projects
Key Phrase Extraction — graph-centrality algorithms (PageRank, degree, centrality) to automatically extract key phrases from conversational data, improving data classification and retrieval.
Sentiment Analysis Benchmarking — evaluated end-to-end, transfer learning, distributed, and explainable AI models across text and emoji data.
Skill-Set Analytics at Scale — big-data graph modeling over large resume datasets to surface skill trends and gaps for talent acquisition.
Cryptographic Algorithm Identification — deep convolutional neural networks to identify encryption algorithms from ciphertext, strengthening cryptographic protocols.
Pothole Detection — real-time deep learning model for road hazard detection, optimized for accuracy and low inference latency.
Graph-Based Anomaly Detection — novel background-knowledge-driven approach to bias substructure discovery toward anomalous patterns, applicable to fraud and risk detection.
Fraud Pattern Detection (Kasios) — Neo4j-based graph analysis to uncover suspicious employee activity patterns in a manufacturing enterprise.
CRADLE: Graph Compression — domain-independent knowledge rules for compressing and visualizing large-scale graph structures. Published, Intelligent Data Analysis Journal (SCI-indexed).
IP Network Analysis — graph-based structural analysis of network infrastructure to detect vulnerabilities.
Graph-Based Cryptographic Ciphers — novel substitution ciphers using matrix algebra and minimum spanning trees over graph structures.
COVID-19 Fake News Detection — contextual graph-based NLP approach to identify and combat misinformation during the pandemic.