# Noel Desmarais

> Software engineer at Maven AGI building applied AI systems for enterprise agents. Focused on LLM orchestration, multi-agent systems, retrieval, agent tooling, and full-stack AI product development.

Noel Desmarais is a software engineer based in the Boston area interested in how applied AI systems can solve real problems. He works at Maven AGI, a venture-backed enterprise AI company building AI agents for customer support. His work focuses on the systems and tooling that help teams design, configure, and ship production agent behavior more effectively. This includes shipping projects like Agent Capabilities—a self-serve environment for extending agent behavior—and broadly sits at the layer where AI capability meets real organizational complexity.

Outside of work, he experiments with agent architectures, proactive behaviors, and the boundaries of what modern AI systems can reliably do. He is especially interested in how agents integrate with legacy systems, operate within real constraints, and become more useful without becoming less predictable.

His projects reflect a consistent interest in turning messy inputs into reliable outputs, whether through receipt extraction, multi-agent planning, or full-stack product design. 

Most of his work sits at the intersection of AI capability and practical execution.

## Current Focus

- Applied AI systems for enterprise agents
- LLM orchestration and multi-agent workflows
- Retrieval, embeddings, and structured AI outputs
- Full-stack AI product development
- Developer tooling for AI systems
- Cloud and serverless infrastructure across AWS, Cloudflare, and Google Cloud

## Selected Projects

- Maven AGI – Agent Capabilities — Built the self-serve tooling environment for Maven AGI customers to extend agent behaviors through custom actions and triggers.
  https://www.mavenagi.com/resources/agent-capabilities

- Quotient — AI-powered receipt scanning and bill splitting
  https://withquotient.com

- Kairos — multi-agent AI weekly planner with calendar integration

- Midwest Road Usage Charge Pilot — production web applications supporting real government pilot programs

- LED Matrix — Raspberry Pi Pico-based LED display recreation
  https://github.com/npd29/Led-Matrix

## Technical Areas

- AI / ML Systems: LLM orchestration, multi-agent systems, RAG, embeddings, semantic search, vector databases, prompt engineering, OCR and document parsing
- Languages: TypeScript, Python, C++, Swift
- Frontend: React, Next.js, Tailwind CSS
- Backend / Data: PostgreSQL, REST APIs, OAuth, schema design
- Cloud / Infrastructure: AWS, Azure, Cloudflare, Google Cloud Run, serverless systems

## Links

- Portfolio: https://desmarais.dev
- LinkedIn: https://www.linkedin.com/in/noeldesmarais/
- GitHub: https://github.com/npd29