Quick bg: started in session-based recsys research (as transformers were taking over), spent some time in product management, then founded a real estate development and investment company that recently sold an eight-figure GDV project in Portugal. We launched a digital R&D arm to build ML- and automation-driven products in the B2B space, which now power the intelligence backend of most of our brokerage partners and save teams more than 150 hours per month. Stepped down as managing partner to find something exciting to build in the tech space. I also enjoy building small tools to tackle niche problems, like keeping my 1K+ browser tabs under control.
Software engineer focused on cloud infrastructure, developer tooling, and AI-powered applications. Career changer; practiced law for several years before switching to engineering full-time.
Most recently I worked at Mastra, an open-source TypeScript framework for AI agents and workflows. I built the PostHog and Datadog observability exporters bridging OpenTelemetry spans with AI analytics, implemented the Azure OpenAI provider with OAuth2 token caching, and resolved critical state propagation issues in Model Context Protocol (MCP) workflows.
I also created Nimbus (https://nimbusnlp.github.io/), an open-source framework that simplifies deploying lightweight NLP models to AWS. TypeScript CLI + AWS CDK to automate provisioning of serverless infrastructure (Lambda via Docker). Takes what used to be 35+ manual steps down to a single command.
I also built RAGCast, a full-stack RAG application that lets users query and get context-aware answers from podcast transcripts using OpenAI, LangChain, and pgvector.
Previously contributed to Mozilla Firefox, enabling extension developers to manually trigger update checks for faster delivery of security patches, with safeguards to prevent keyboard shortcut conflicts.
I don't mind doing technical tests but I hate technical interviews (where people have to watch me type) really if you want to know what kind of code I can produce look at my github repos, there is easily 100-200k lines of code
