Engineered a frontend dashboard and REST API using Next.js to centralize content creation and management with an AI-first approach. The platform includes a built-in markdown editor powered by TipTap, leveraging TipTap's AI features to capture and replicate user-specific writing styles, significantly accelerating content development. It also implements custom agentic web scrapers for knowledge retrieval and resource aggregation, streamlining the content generation process. Integrated with APIs from major blog providers (Medium, LinkedIn, Dev.to) to enable efficient distribution across multiple platforms.

Built a CLI tool that lets me control my computer with voice commands using OpenAI’s Real-Time API. It’s tailored for coding workflows like running shell commands, web scraping documentation, and saving it in an AI-friendly markdown format, as well as file and directory creation. Designed with a modular structure, it’s ready for adding more tools in future iterations. I also integrated the audio buffer mechanism from my Twilio Phone Assistant, ensuring smooth and responsive voice interactions, even when I interrupt the AI mid-task.

A landing page for my AI consultancy business that I'm launching with a fellow engineer, built from scratch within a couple of hours using Next.js, Tailwind CSS, Framer Motion, and Rombo's Tailwind motion library.

Developed a Twilio Phone Assistant powered by OpenAI's Real-Time Assistant API to deliver immediate, context-aware responses during live phone conversations. The system is built with FastAPI to handle the API layer and Twilio for seamless call management. Implemented an integrated audio buffer to enhance interaction quality and effectively manage interruptions, improving the overall user experience.

Developed an automated pipeline for generating tailored slide deck content from meeting transcripts and AI-generated notes (via Circleback.ai). Leveraged LlamaIndex with Pydantic output parsing, implemented prior to OpenAI's structured output response model. Built with FastAPI and Jinja2 templates for prompt templating, the system enables efficient slide content generation. The use of Jinja2 allows for highly flexible prompt injection, offering extensive customization to adapt outputs to specific requirements.

A FastAPI Microservices Architecture leveraging Redis for messaging, enabling dynamic public and private knowledge repositories for Retrieval-Augmented Generation workflows. Integrates Pinecone for embeddings and Neo4j for graph-based data, with a scalable CI/CD pipeline for seamless deployment of containerized microservices.

A challenge that required me to build a basic package manager in 2 hours with Node.js and implement two commands: add (to add a package) and install (to install all packages). Decided to use a depth-first-search traversal algorithm for dependency management.

An illustrative SaaS landing page I built to explore Framer Motion scroll animations and demonstrate smooth user interaction concepts.