TypeScript at scale. AI in production. Shipped on my own infra.

TypeScript · React · Node · Python · MCP · Docker · CI/CD

I ship typed React and Node services, integrate LLMs through Model Context Protocol, and run the infra end-to-end on Docker, VPS, and Dokploy. 7+ years in production. Open source on GitHub.

Profile

Three things, deep: typed full-stack systems, AI integrations through Model Context Protocol, and the ops layer that keeps it all running.

Strict TypeScript on both ends — React 19 on the front, Node and FastAPI services on the back. Type errors, tests, and deploys are part of the product, not afterthoughts.

I build and ship MCP servers that expose real tooling to Claude and other LLMs. NotebookLM MCP — 32 tools, open source — is one of them.

Infra runs on Docker, Dokploy, Portainer, and VPS. CI/CD wired in GitHub Actions. Code, AI plumbing, and deploys, end-to-end.

Years in production
7+
Systems shipped
30+
MCP tools open-sourced
32
TypeScript strict
100%

Stack

What I reach for when building typed services, AI integrations, and the infra behind them.

  • TypeScript

    • TypeScript (strict)
    • React 19
    • Next.js
    • Node.js
    • NestJS
    • Express
    • Zod
    • Vite
    • Vitest

  • AI / MCP

    • Model Context Protocol
    • Claude API
    • Anthropic SDK
    • Agent design
    • Prompt engineering
    • RAG patterns
    • NotebookLM

  • Ops

    • Docker
    • Docker Compose
    • Dokploy
    • Portainer
    • VPS (Linux)
    • GitHub Actions
    • CI/CD
    • Nginx

  • Backend

    • Python
    • FastAPI
    • Django
    • REST APIs
    • GraphQL
    • Alembic
    • Auth (JWT, OAuth)

  • Data

    • PostgreSQL
    • Redis
    • SQL tuning
    • Migrations
    • Caching

  • Practice

    • Strict typing
    • TDD
    • Code review
    • Clean architecture
    • Observability

Selected work

Production systems I designed, shipped, or led. Open a card for the case study.

AI / MCP2026

NotebookLM MCP Server

Creator and maintainer

Problem: LLM agents had no programmatic surface into NotebookLM. Solution: a production-grade Model Context Protocol server exposing 32 typed tools covering notebooks, sources, research, and content generation. Outcome: open-source reference implementation, used as the canonical NotebookLM MCP integration.

  • 32 MCP tools — full NotebookLM surface area
  • Strict TypeScript, typed contracts at every boundary
  • Open source, MIT, actively maintained
  • Reference impl for NotebookLM × LLM agents
  • TypeScript
  • MCP
  • Node.js
  • NotebookLM
View repository
AI / Media2025

Transarr

Architect and lead engineer

Problem: paid subtitle SaaS hits limits fast and never owns your pipeline. Solution: full stack media service — faster-whisper large-v3 transcription, Meta NLLB-200 translation across 200+ languages, React UI on a Flask backend, FFmpeg under the hood. Outcome: zero per-minute cost, full control, runs on my VPS.

  • faster-whisper large-v3 transcription
  • Meta NLLB-200 translation, 200+ languages
  • React + Flask + FFmpeg, fully Dockerized
  • Self-hosted on VPS, monitored end-to-end
  • Python
  • Flask
  • React
  • TypeScript
  • Docker
  • FFmpeg
Private — under NDA / in development
Full Stack2025

StreamFlow

Contributor

Streaming service split clean: FastAPI controllers/services/repository for management and transcoding, React frontend for playback with multi-language subtitles. Containerized with Docker, deployed on personal VPS.

  • Adaptive streaming via FastAPI
  • Multi-language subtitle pipeline
  • Clean controller/service/repository split
  • VPS-deployed with Docker
  • FastAPI
  • React
  • Python
  • Docker
  • FFmpeg
View repository
Fintech / Backend2023 – present

Hanclym Trading

Technical co-founder

Problem: needed an opinionated trading platform with portfolio, orders, and market analysis under our own roof. Solution: FastAPI core with a typed REST API, Alembic migrations under version control, Redis cache for low-latency reads, PostgreSQL tuned for read patterns, microservice-shaped, Dockerized. Outcome: production system handling our own trading flow.

  • Typed REST API across the trading surface
  • Alembic migrations under version control
  • Redis read-through cache for hot paths
  • Microservice-shaped, Docker-deployed on VPS
  • Python
  • FastAPI
  • PostgreSQL
  • Redis
  • Alembic
  • Docker
Private — under NDA / in development
Dev Tools2022 – present

Hanclynvim

Creator and maintainer

Neovim config built on LazyVim, tuned for full-stack engineering. GitHub Copilot integrated, LSP across TypeScript, Python, Lua, debugging wired in, 15+ plugins picked for signal-to-noise.

  • LazyVim base with strict lazy-loading
  • Copilot integrated end-to-end
  • LSP for TypeScript / Python / Lua
  • Actively maintained — used daily
  • Lua
  • Neovim
  • LazyVim
  • Copilot
  • LSP
View repository
Fintech2019 – 2021

SideUp

Full stack engineer

Enterprise fintech: payment management, international transfers, banking integrations. Worked across backend (Node) and frontend (React) on the core system and gateway integrations.

  • International payment gateway integrations
  • Core transfer and payment modules
  • Node + React stack
  • Agile team delivery
  • React
  • Node.js
  • PostgreSQL
  • REST APIs
  • Docker
Private — under NDA / in development
Blockchain2018

Skycoin

Open-source contributor

Contributions to Skycoin, a third-generation cryptocurrency with a decentralized internet stack. Authored protocol documentation, transaction guides, and developer tooling around a Go codebase.

  • Technical protocol documentation
  • Transaction guides for developers
  • OSS contributor — 1K+ stars repo
  • Go + blockchain ecosystem
  • Go
  • Blockchain
  • Documentation
View repository

Reach out

GitHub for code. Email for proposals. LinkedIn for everything else.

Direct lines

Email[email protected]LinkedInMaykel Arias TorresGitHub@m4yk3ldev

Elsewhere

Working on something specific?

Send context, goal, and timeline. I reply with an opinionated plan or a clear no within 24h.

EmailGitHub