Software & AI Engineer: backend systems, LLM agents, and voice AI
I'm a software and AI engineer who builds backend systems and LLM-powered products. Lately that's meant a voice AI pipeline that answers real customer calls and AI agents that can safely handle payments and reservations. I enjoy the parts that make these systems dependable: the data pipelines, the testing, and the deployments underneath.
I build the voice AI pipeline and multi-agent platform that businesses use for customer calls, including approval checks that let agents safely take payments and make reservations. The voice pipeline handles 50K+ customer calls a month.
Took a document retrieval app from first prototype to production (FastAPI, TypeScript, LangChain, and Pinecone on AWS ECS), and built a SageMaker framework for fine-tuning small language and embedding models. The retrieval app ran at 99.99% uptime.
Built backend APIs and data pipelines with Flask, PostgreSQL, RabbitMQ, and Airflow, plus an internal KPI tracking tool for the sales and logistics teams. Moving the Flask APIs onto SQLAlchemy cut p95 latency from 450ms to 150ms.
A real-time messaging app with personal and group chats.
Built a React front end on Laravel, with Laravel Reverb WebSockets pushing new messages and notifications to open chats instead of polling. It supports Markdown, emoji, image and document attachments, and group management, backed by MySQL.
View on GitHub (engaging)An email app that uses AI agents to search, summarize, and reply to your inbox.
Co-designed a LangGraph multi-agent system with prompt-correction, retrieval, summarization, and reply agents. Implemented the multi-agent setup, email and attachment embeddings for semantic search, the FastAPI chatbot endpoint, Microsoft sign-in, and on-device email categorization with Phi-3.
View on GitHub (Inboxify)A small language model fine-tuned to answer investment questions in George Soros's style.
Fine-tuned Google's FLAN-T5-Base on a custom set of about 600 question-and-answer pairs covering psychology, risk management, adaptability, timing, and strategy, then evaluated it with ROUGE, METEOR, and BERTScore. An educational project, not financial advice.
View on GitHub (SorosAdvisor)A campus life tracker that brings Northeastern student events into one place.
Co-developed a full-stack web app with event search and category filters, RSVPs, bookmarks, and reminders, plus accounts for students, organizers, and admins.
View on GitHub (huskytrack)A small tweeting app that runs as three containerized microservices.
Built a React front end, a Flask API, and a MongoDB database, each in its own Docker container, and deployed them to AWS ECS behind load balancers with auto scaling. AWS CodeBuild builds the images on every push, and Terraform defines the infrastructure.
View on GitHub (twtr)