Project Portfolio

Sharon Marfatia

UBC Computer Science · Available May 2026 · AWS Certified Solutions Architect Associate

Multi-Agent System · Quantum Computing
Quantum Materials Assistant
May 2025 – Dec 2025
Strands Agents SDK Amazon Bedrock Amazon Braket MCP Materials Project MCP Claude Sonnet 4.5 Qiskit Python
60% reduction in modeling effort
300+ simulations run
140K+ compounds accessible

A multi-agent platform enabling natural-language generation of runnable quantum simulation code, deployed for physicists at Stewart Blusson Quantum Matter Institute and The Quantum Insider. Orchestrates DFT, structure analysis, quantum theory, and agentic-loop workflows via Strands Agents SDK with Amazon Braket MCP and Materials Project MCP integration, eliminating manual Python scripting for researchers.

GenAI Platform · Healthcare Education
Empathetic Communication Trainer
May 2025 – Dec 2025
Nova Pro Nova Sonic RAG Pipeline Amazon RDS (pgvector) WebSockets ECS Fargate Python
100+ pharmacy students
1st automated empathy scoring
UBC formally adopted

A GenAI platform formally adopted into UBC's pharmacy curriculum, simulating AI patients via chat and voice to train empathy skills previously unmeasured. Integrates speech-to-speech AI via Nova Sonic, a RAG pipeline for context-aware patient scenarios, and instructor dashboards tracking student empathy performance across structured evaluation dimensions.

Distributed Systems · Drug Discovery Infrastructure
Protein Structure Analysis Pipeline
May 2024 – Dec 2024
Kubernetes AWS Autoscaling Queue Orchestration Python Docker
97%reduction in analysis time
1,000+protein structures processed
15 min → 30sper structure

A distributed containerized data-extraction pipeline on Kubernetes (AWS) with autoscaling and queue orchestration to parallelize workloads across 1,000+ protein structures. Enabled computational biologists to submit batch CSV workloads overnight and receive structured protein feature extracts by morning, replacing a fully manual MOE GUI workflow.

ML Engineering · Protein Thermostability Prediction
ESM-2 Thermostability Models
May 2024 – Dec 2024
ESM-2 Transformer TensorFlow PyCaret XGBoost Python Drug Discovery
1.89°CMAE achieved
33%better than benchmark
2.8–3.0°Cindustry benchmark

State-of-the-art protein thermostability prediction using ESM-2 transformer models, outperforming the industry benchmark of 2.8–3.0°C MAE by 33%. Enabled the computational biology team to screen drug candidates for aggregation and solubility risks without costly lab testing, reducing dependence on wet lab validation in early-stage drug discovery.

Data Engineering · Sustainability · Published Research
GHG Eco-Label Classification System
May 2023 – May 2024
Azure ML Studio Python Classification Pipeline Data Cleaning GHG Emissions UBC Food Services
30+UBC venues deployed
5,000+daily diners reached
6,000+food products standardized

Deployed GHG eco-labels across 30+ UBC restaurants and cafes reaching 5,000+ daily diners by building a classification pipeline that extracted XML dining data, standardized 6,000+ food products, and categorized meals into red/yellow/green emissions tiers. Research formally published in the UBC Open Library in collaboration with UBC Sustainability and UBC Food Services.

Chrome Extension · Google Built-in AI Challenge
PageClarity
Oct 2025 – Nov 2025
Chrome Built-in AI Gemini 2.5 Pro Manifest V3 JavaScript ES6+ Four-Tier Architecture
5 Chrome Built-in AI APIs
100% CSP-restricted site support

A hybrid AI Chrome extension enabling in-tab reading assistance on all sites including CSP-restricted pages. Uses a four-tier routing architecture across 5 Chrome Built-in AI APIs (Summarizer, Rewriter, Translator, Proofreader, Prompt) with Gemini 2.5 Pro fallback, eliminating tab-switching for summarization, translation, rewriting, proofreading, and Q&A.

Autonomous AI Agent · AWS AI Agent Global Hackathon
FollowUpSync
Sep 2025 – Oct 2025
Amazon Bedrock Nova Micro FastAPI MCP Slack API Notion API Amazon S3 Python
15+ min saved per meeting
<30s transcript to task delivery

An autonomous AI agent that eliminates lost meeting action items by parsing transcripts via Amazon Bedrock Nova Micro and auto-delivering structured tasks to Slack channels with tagged recipients and Notion databases via FastAPI MCP servers. Built for the AWS AI Agent Global Hackathon, stores structured meeting summaries in S3 for audit and retrieval.

Amazon Web Services
AWS Certified Solutions Architect – Associate
Validates expertise in designing scalable, secure, and resilient cloud architectures

Applied across distributed Kubernetes pipelines, multi-agent GenAI platforms, and RAG architectures at AWS Cloud Innovation Center and Zymeworks Inc. Covers cloud architecture design, AWS service selection, security best practices, and cost optimization across EC2, ECS, Lambda, S3, SageMaker, Bedrock, and CloudWatch.