Matias – Python, Typescript, AI agent development

LLM-powered agent capable of autonomously controlling an entire mobile operating system beyond the browser.

• Contributed to multimodal, OS-level automation capabilities;
• Developed agent backend and infrastructure components as part of the core platform team.

The world’s first LLM-powered agent capable of autonomously navigating and interacting with a web browser.

• Led the end-to-end V2 platform rebuild;
• Designed and implemented a modular architecture;
• Developed agent planning and reasoning systems;
• Implemented parallel multi-LLM orchestration;
• Increased reliable agent execution from 6–7 to 40–50 consecutive steps;
• Led the development of the "Teach Me" feature;
• Helped define engineering standards and hiring practices.

Provided pro bono technical advisory and architecture consulting for early-stage startups in Bangladesh and Saudi Arabia.

• Advised founding teams on applying emerging LLM techniques, including chain-of-thought prompting, few-shot learning, RAG, and agentic architectures, before widely adopted production patterns had been established;
• Helped clients clarify product goals and translate experimental research concepts into practical implementation strategies;
• Bridged cutting-edge LLM research and product development by turning new techniques into actionable technical approaches;
• Delivered technical leadership that resulted in a full-time offer to join MultiOn and work on its agent platform.

Delivered engineering solutions for clients across the US, Germany, and the UK, including a London-based cryptocurrency startup.

• Built LangChain integrations and RAG systems for early-stage startups at a time when production patterns for LLM applications were still emerging, translating experimental techniques into working software;
• Designed and developed a multi-LLM cryptocurrency research agent that combined social signals with time-windowed market data to generate long- and short-term outlooks for selected tickers;
• Built pipelines that ingested posts from major crypto influencers, summarized and classified them using LLMs, and combined sentiment analysis with technical market calculations;
• Implemented an early voice-driven customer service system using ElevenLabs and speech transcription models, enabling users to complete tasks in real time through natural phone conversations.

Private networking platform for a global entrepreneurial community, enabling members to connect, collaborate, and access mentorship

• Developed AI-powered features for a private networking platform;
• Proposed and developed one of the first production chatbots deployed in Argentina;
• Built custom vector database integrations;
• Implemented data segmentation to overcome the 8K token context limit;
• Contributed to the LangChain open-source framework;
• Led feature development, bug fixes, and production deployments as Tech Lead.