● Built a multimodal interruption detection model by fusing BERT, librosa, and MediaPipe features in PyTorch, achieving 84% F1 accuracy
● Synced transcripts, speakers, and frames across 9K+ debate segments using ffmpeg, Whisper, and pyannote-audio to enable supervised
multimodal training
● Exported model to ONNX, reducing inference latency by 42% and memory by 60%, enabling real-time interruption detection during
live-streamed political debates
● Developed supporting infrastructure with FastAPI, PostgreSQL, and Docker, integrating privacy-compliant data handling and CI/CD via
GitHub Actions

● Built and deployed an AI-powered learning iOS app using React Native, FastAPI, and PostgreSQL, reducing onboarding time for seniors
by 60% through native navigation and App Store delivery
● Integrated Whisper and ChatGPT via OpenAI APIs, enabling real-time voice control and hands-free assistance tailored for older users
● Containerized REST APIs with Docker and deployed to AWS EC2, maintaining 99.9% uptime and performance across iOS devices
● Implemented Firebase Auth, OTA updates, and Apple-focused CI/CD pipelines, ensuring secure sessions,, and smooth App Store reviews

Western University London, ON
● Awarded $10,000 to build a cross-platform AI agent app using React Native, Firebase, GPT-4, LangChain, and FAISS, enabling real-time
news-to-research mapping on both iOS and web without requiring academic credentials
● Fine-tuned DistilBERT on AWS SageMaker, improving match accuracy by 85% for relevance-ranked retrieval
● Automated training and deployment with Docker, MLflow, and GitHub Actions, enabling continuous retraining with zero manual ops

● Designed encoder-based AI agents using FastAPI, LangChain, OpenAI APIs, and Firebase, classifying over 10M logs/day to power
real-time personalization and privacy-compliant decision pipelines
● Built and containerized a scalable microservice architecture with Docker, PostgreSQL, and Kafka, enabling 12+ services to run
concurrently under live production workloads
● Automated CI/CD and MLOps workflows using MLflow, Jenkins, and GitHub Actions, cutting deployment and QA cycles by 50%
● Integrated Supabase, Sentry, Datadog, to streamline analytics instrumentation, and secure model testing during daily rollouts

● Built a Q&A agent with React Native, Firebase, and GPT-4, serving 1,500+ weekly users with >95% accuracy and 1–3s response times.
● Integrated Supabase and Prisma for live session tracking and structured user prompts, powering persistent agent memory
● Shipped a search interface using Typesense and PostgreSQL, improving document relevance and reducing user drop-off by 40% through
instant mobile responsiveness
● Fine-tuned LLMs on SageMaker using PyTorch and deployed via Docker/Kubernetes, reducing inference errors by 33%

● Engineering a book marketing tool with Node.js, FastAPI, PostgreSQL, and Docker, deployed on AWS EC2 with zero manual intervention
● Built async outreach pipelines using headless browsers, custom schedulers, and rate-limit bypassing to enable 1,000+ targeted DMs/day
● Used GPT-4o with scraped profile data to generate personalized outreach, increasing reply rates by 43%

● Built autonomous content agents using GPT-3.5-turbo and DistilBERT to generate SEO-optimized blogs, scaling to 80K+ monthly
readers, $2.8K/month net profit, and a mid–five-figure exit
● Developed a full-stack backend with FastAPI, Node.js, Firebase, PostgreSQL, and Celery, automating content workflows with secure,
GDPR-compliant data handling
● Deployed scraping agents using Playwright and Puppeteer to extract trends from Reddit and YouTube, increasing CTR by over 3×
● Implemented CI/CD with GitHub Actions and Docker, integrated monitoring via Sentry, and automated prompt tuning with Typesense to
maintain top Google rankings