hey, I'm Nico 👋
Software engineer building at the intersection of AI and the browser. Based in Chile.
I've been coding since 2000, started with the early web, went through PHP, jQuery, backbone, the whole ride. These days I focus on AI agent architecture, LLM integration patterns, and making intelligent systems work natively in web environments.
what I'm working on
I spend most of my time thinking about how AI agents should communicate, how to bring ML capabilities into the browser, and how to design systems that are composable and protocol-driven.
Current focus areas:
- Agent protocols & architecture - designing how agents talk to each other and to the world
- Browser-native AI - running LLM-powered tools inside the browser, not just calling APIs
- NLP in the frontend - tokenization, embeddings, semantic search, all client-side
- Web app architecture - clean patterns, domain-driven, influenced by Fowler and the classics
projects
| Project | What it does |
|---|---|
| nexu | Chat with your codebase. AST-aware chunking + graph expansion for precise code retrieval |
| browser-agent-kit | Toolkit for building AI agents that operate inside the browser. DOM analysis, smart selection, action planning |
| agent-protocol-ts | TypeScript implementation of the Agent Protocol spec with middleware system |
| SnapLite | Chrome extension. One-click screenshots compressed for AI. No cloud, no bloat |
| prompt-graph | Compose LLM prompts as directed acyclic graphs. Functional, composable, testable |
| llm-stream-parser | Streaming JSON parser for LLM responses. Handles partial output, recovers incomplete JSON |
| embedding-viz | Interactive embedding space visualization. PCA/t-SNE projection, Canvas rendering, pan & zoom |
| tokenizer-arena | Compare how different LLM tokenizers break down the same text, side by side |
| react-graphie | React library for interactive graph visualizations using D3.js |
background
25 years writing software. I started with HTML in a cybercafé, got obsessed with making things interactive, and never stopped. Over the years: vanilla JS, jQuery, Angular, React, whatever comes next. Always JavaScript at heart, always the browser as the platform.
In recent years I got deep into software architecture, domain modeling, event-driven systems, the Martin Fowler school of thought. Then LLMs happened and everything clicked: the web is the best delivery mechanism for AI, and the browser is the most powerful runtime we have.
Now I work on the infrastructure layer, the protocols, patterns, and tools that make AI agents actually useful in production web applications.
philosophy
- The browser is underrated as an AI runtime
- Good architecture matters more than good prompts
- Protocols > frameworks, standards outlast hype
- Ship small things that work, not big things that don't
- If you can't explain it simply, you don't understand it
reach me
- X: @nicolascine
- Chile 🇨🇱