Works

Projects

GenAI Virtual Assistants for Contact Centre Operations

Leading GenAI transformation at Manulife across USA (John Hancock) and Canada, building AI agents for customer service reps using Azure AI Search, OpenAI, Databricks, and LangGraph.

LLMs AI Agents LangGraph Azure

Multi-Agent AI Risk & Privacy Assessment

Built a multi-agent prototype with CrewAI and Cohere Command R+ to automate AI risk assessment by identifying ethical and data privacy concerns. Hackathon placer at Cohere Build Day.

Multi-Agent CrewAI Cohere Responsible AI

RAG Knowledge Base Search at GCash

Leveraged RAG, prompt engineering, and OpenAI GPT-3.5/4o to improve company knowledge base search accuracy by 24% and speed by 73% based on A/B testing.

RAG OpenAI A/B Testing

Text-to-SQL Chatbot

Developed a text-to-SQL chatbot POC with LangChain and Azure OpenAI GPT-3.5, achieving 81% SQL generation and 87% Q&A accuracies. Secured deployment into expense tracker product as beta feature.

LangChain Text-to-SQL Azure OpenAI

Privacy Impact Assessment RAG Agent

Built an RAG agent using Azure OpenAI GPT-4o and Cohere Rerank to expedite Privacy Impact Assessments, achieving 86% Q&A accuracy and elevating data privacy awareness organization-wide.

RAG Cohere Rerank Data Privacy

HR Survey Topic Modeling & Summarization

Optimized HR survey analysis by 90% using NLTK, Gensim, TopicBERT for topic modeling and Anthropic Claude for summarization, saving 270 man-hours.

Topic Modeling Anthropic Claude NLTK

Multi-PDF Conversational App with Open Source LLMs

Deployed a Streamlit app for multi-PDF uploads with RAG-based conversational Q&A powered by Mistral-7B-instruct, LangChain, and Annoy Vector DB.

Mistral RAG Streamlit

LLM Fine-Tuning for Speaking Style Emulation

Fine-tuned Gemma-2b-instruct to mimic a specific speaking style, integrated open-source voice cloning for conversational interactions via Streamlit chat interface.

Fine-Tuning Gemma Voice Cloning

Credit Lending Fairness Audit

Conducted fairness audit on credit lending models using Fairlearn, analyzing demographic parity, equalized odds, and equal opportunity across age, marital status, and gender.

Fairlearn Responsible AI Model Fairness

Community Structure of ASEAN Corporate Network

Quantified Trans-ASEAN integration using interlocking board directorates and network science tools to identify market consolidation and power dynamics.

Network Science Graph Analytics

SDG News Articles Topic Modeling

Identified gaps in global SDG achievement using LDA topic modeling on 1,000+ news articles from the SDG Knowledge Hub.

NLP Topic Modeling LDA

Pandemic Pulse: Global Trends via GDELT

Processed 182GB of GDELT Global Knowledge Graph data using distributed LDA and PySpark to construct topic models tracing pandemic trends over time.

Big Data PySpark LDA

Emotion Distribution in News Articles

Developed ML models to classify emotion intensity in 17,000+ Rappler articles using Word2Vec embeddings and ensemble tree methods.

Machine Learning NLP Word2Vec

COVID-19 Government Response Analysis

Applied agglomerative and hierarchical clustering on presidential broadcast transcripts to track the evolution of pandemic response priorities.

Clustering NLP Data Mining

Property Price Prediction at Unionbank

Led development of a LightGBM model for property price prediction using web-scraped data and OpenStreetMap geospatial features, achieving 20-58% MAPE.

LightGBM Geospatial ML

Virtual Try-On with Generative Neural Networks

Built a virtual clothing try-on system using generative neural networks for realistic image synthesis.

Deep Learning Computer Vision GANs

DeepRec: Hybrid Recommender System

Built a hybrid recommender system combining collaborative filtering with deep learning for improved recommendation quality.

Deep Learning RecSys

Vaccination Planning via Network Centrality

In-silico simulation of vaccine strategies using centrality measurements to optimize vaccination planning.

Network Science Simulation