AI & Agents
- LLMs: OpenAI · Azure OpenAI · Ollama (local) · OpenRouter · Anthropic
- Vector DBs: Qdrant · Chroma · Pinecone · pgvector
- RAG pipelines · Autonomous agents · LangChain · Prompt engineering · Evals & guardrails
- N8N · Honcho
Hi, my name is
David Buitrago.
I build systems that think._
I'm an AI Systems Engineer based in Madrid, Spain, specializing in LLM integrations, agentic automation, and internal platforms that remove operational overhead. Currently Head of IT at Qualitas Funds, where I lead the technology architecture and build AI-powered tools for the investment management industry.
01.About Me
I'm a software engineer and AI systems builder with over 10 years of experience designing and shipping software that runs — not just demos that look good in a presentation. I started my career in enterprise software and digital transformation, and over time I've shifted my focus entirely toward building intelligent systems: autonomous agents, LLM pipelines, and the internal platforms that make teams move faster.
My background is in .NET, SharePoint and enterprise integrations, but today my stack lives at the intersection of AI and infrastructure: N8N for orchestration, LLMs for reasoning — from open-source models via Ollama to cloud APIs through OpenRouter and Azure OpenAI — and vector databases for semantic search. I work with whatever gets the job done reliably in production. I care about observability, maintainability, and systems that don't require a manual to operate.
At Qualitas Funds, I lead the full technology scope: cloud infrastructure, servers, networking, databases, and a development team building AI-powered tools for the investment management industry — autonomous agents, data pipelines, and workflow automation for complex financial processes. I believe the best engineers today are the ones who can move fluidly between infrastructure, AI, and product thinking.
Outside of work, I run ADSIS — a B2B SaaS platform I built from scratch that uses AI to help digital marketers generate high-performance ad creatives. I also run a homelab with Docker-based self-hosted infrastructure, experiment with local AI agents, and occasionally try to teach my Anki Vector robot new tricks.
02.Skills
AI, infrastructure, backend development, and technical leadership.
Cloud & Infrastructure
- Azure (App Services · SQL · Storage · Networking · Entra ID · DevOps)
- Docker · Tailscale · GitHub Actions · CI/CD
- Server & network management · Vendor management
Backend & Development
- Python · .NET Core / C# · Azure SQL · SQL Server · REST APIs
- React · TypeScript · Next.js
Architecture & Leadership
- Solution architecture · Technical leadership · Vendor management
- Team mentoring · Git branching strategies · Technical documentation
03.Where I’ve Worked
Head of IT · Tech Lead @ Qualitas Funds
Mar 2021 – Present · 4+ years · Madrid, Spain
- Lead the technology architecture across all company applications, defining standards for infrastructure, development workflows, and AI integration. Oversee servers, networking, databases, web apps, and Azure services across the organization.
- Designed and deployed an agentic automation system using N8N, Azure OpenAI, and Azure SQL to automate complex financial workflows, including commission calculations, investor onboarding processes, and regulatory reporting pipelines.
- Implemented a RAG pipeline with vector search and Azure OpenAI that allows the team to search and reason over large volumes of internal documentation and financial reports.
- Manage vendor relationships and technology procurement — software licenses, cloud infrastructure, external providers — and lead a team of developers through mentoring, code review, and architectural guidance.
Stack: Azure OpenAI · N8N · Vector databases · .NET Core · Azure SQL · React · TypeScript · Docker · Azure DevOps · Azure (App Services · Networking · Storage · Entra ID)
SharePoint Development Engineer @ Syntax Informática
2019 – Mar 2021 · 2 years · Madrid, Spain
- Delivered custom SharePoint solutions and corporate intranet platforms for enterprise clients across multiple industries, including full digital transformation projects.
- Designed software architectures for complex SharePoint customizations, integrating with Microsoft 365 services, custom APIs, and legacy enterprise systems.
- Led the technical delivery of intranet projects from requirements through deployment, working directly with clients to align technology choices with business needs.
Stack: SharePoint · Microsoft 365 · .NET · SPFx · Azure AD · Power Platform
Software Development Engineer @ Intergrupo
Oct 2017 – Mar 2019 · 1.5 years
- Designed and developed software architectures for multiple enterprise projects, primarily within the Microsoft ecosystem (.NET, SharePoint, SQL Server).
- Built web applications using React for client-facing interfaces, combined with .NET backends and SQL Server data layers.
- Collaborated on architecture definitions for projects of varying scale, from departmental tools to company-wide platforms.
Stack: .NET · SharePoint · SQL Server · React · REST APIs
Technical Lead · Digital Transformation Consultant @ Conocimiento Corporativo
Mar 2014 – Oct 2017 · 3.5 years
- Advised enterprise clients on technology strategy and digital transformation, helping them select and implement the right platforms for their business needs — predominantly Microsoft 365 and SharePoint ecosystems.
- Led technical projects from discovery through delivery, working closely with stakeholders to translate business requirements into architectural decisions.
- Established myself as a trusted technical reference for clients navigating the transition from on-premise infrastructure to cloud-based collaboration platforms.
Stack: SharePoint · Office 365 · .NET · SQL Server · Solution architecture
04.Some Things I’ve Built
Featured Project
ADSIS
adsis.ai +AI-powered ad creative platform for performance marketers
ADSIS is a B2B SaaS platform I designed and built from scratch that helps digital marketers create high-performance ad creatives using AI. Users research their target audience (ICP) through an AI pipeline, then generate image and copy variations ready to deploy on social media campaigns.
The platform runs a multi-step research pipeline: 7 parallel Perplexity queries, relevance filtering via GPT-4o, embedding-based deduplication, and final ICP synthesis — all triggered from a single canvas interface. Image generation runs through OpenRouter (Gemini), copy through GPT-4o, with a credit system that tracks every generation atomically.
Built and deployed entirely by me: architecture, backend, frontend, infrastructure, CI/CD. Running in production on a self-managed VPS with Docker, nginx reverse proxy, Let's Encrypt, and Watchtower for zero-downtime deploys.
- ASP.NET Core 9
- React 19
- TypeScript
- Supabase (Postgres + Auth)
- Hangfire
- OpenRouter
- Perplexity API
- Docker
- GitHub Actions
- Tailwind CSS v4
- Multi-tenant B2B SaaS with tenant-scoped data isolation (X-Tenant-Id middleware)
- AI research pipeline: Perplexity → GPT-4o relevance filter → embedding dedup → ICP JSON synthesis, fully async via Hangfire background workers
- Image and copy generation through OpenRouter (GPT-4o + Gemini)
Featured Project
agent-memory-starter
Persistent memory boilerplate for AI agents
A production-ready starter kit for building AI agents with persistent, cross-session memory. Built on top of N8N for workflow orchestration and Honcho for user memory management, it solves one of the most underestimated problems in production AI systems: making agents remember context across conversations without reinventing the wheel.
The kit ships with a working N8N workflow, a Docker Compose setup, and a clear architectural pattern that can be adapted to any LLM provider. Originally built as the backbone for my personal AI assistant, Donna.
- N8N
- Honcho
- OpenAI API
- Docker
- Python
- Plug-and-play N8N workflow with Honcho memory integration
- Docker Compose for one-command local setup
- Works with any LLM provider: OpenAI, Azure OpenAI, Ollama, OpenRouter
Featured Project
n8n-workflows-kit
Production-ready workflow collection for AI automation
A curated library of N8N workflows for common AI automation patterns: RAG pipelines, document ingestion, semantic search with Qdrant, webhook-triggered agents, and notification systems. Each workflow is documented with its architecture diagram, setup instructions, and configuration notes.
Built from real workflows used in production, cleaned up and packaged so anyone can deploy them in minutes rather than hours.
- N8N
- Qdrant / pgvector
- OpenAI / Ollama / OpenRouter
- Docker
- Webhooks
- RAG pipeline: document ingestion → chunking → embedding → Qdrant storage → Q&A
- Semantic search workflow with Qdrant and OpenAI embeddings
- Webhook-triggered agent with tool use and structured output
Featured Project
llm-financial-tools
LLM toolkit for financial data analysis
A Python toolkit for financial analysis powered by LLMs. Uses public data sources (Yahoo Finance, FRED) to demonstrate how language models can be applied to real financial analysis tasks: portfolio summaries, macroeconomic context, earnings report interpretation, and anomaly detection in time-series data.
Designed with production constraints in mind: latency budgets, cost-per-query tracking, and structured output validation so the results are actually usable downstream.
- Python
- LangChain
- OpenAI / Azure OpenAI / OpenRouter
- Pydantic
- yfinance
- FRED API
- Financial data connectors: Yahoo Finance, FRED (Federal Reserve Economic Data)
- LLM-powered analysis: portfolio summary, sector commentary, earnings interpretation
- Structured output with Pydantic validation — no hallucinated numbers
Featured Project
obsidian-ai-assistant
Semantic search and AI Q&A for your Obsidian vault
An AI assistant for Obsidian vaults that goes beyond keyword search. Documents are ingested, chunked, and embedded into a Qdrant vector database via an N8N pipeline. A conversational interface lets you ask questions across your entire knowledge base and get answers grounded in your own notes — with source references.
Built for my personal knowledge management workflow, where I maintain a vault of several thousand notes synced to GitHub.
- N8N
- Qdrant
- OpenAI / Ollama
- Docker
- Python
- Obsidian
- Automatic vault ingestion via N8N workflow (file watcher → chunker → embedder)
- Semantic search powered by a vector database (Qdrant by default, swappable) and OpenAI embeddings
- Conversational Q&A with source attribution
05.Other Noteworthy Projects
portfolio-v2
This portfolio — built with Next.js and Tailwind, inspired by Brittany Chiang's v4
- Next.js
- Tailwind
- TypeScript
vector-local-ai
Experiment reviving an Anki Vector robot with local AI using Wire-Pod and Ollama
- Python
- Ollama
- Wire-Pod
- Docker
donna
Personal AI assistant with persistent memory, voice interface, and Obsidian integration
- OpenClaw
- ElevenLabs
- N8N
- Honcho
06.Some Things I’ve Written
How I added persistent memory to my AI agent (without reinventing the wheel)
Most AI agent tutorials ignore memory. They show you how to build a chatbot, but the moment you close the tab, the conversation is gone. This post covers how I integrated Honcho with N8N to give my personal assistant real persistent memory across sessions — what worked, what didn't, and the architectural decisions I'd make differently today.
RAG in production vs RAG in demos: what nobody tells you
Retrieval-Augmented Generation looks deceptively simple in tutorials. In production, it's a different story. After building RAG pipelines for a financial firm and for my own knowledge base, I've collected a list of things that break in ways the docs don't cover — chunking strategies, embedding drift, retrieval recall vs precision tradeoffs, and why your vector database choice matters more than your LLM choice.
My 5 most useful N8N workflows for AI automation
After two years of building automation with N8N, these are the five workflows I reach for again and again: a document ingestion pipeline, a webhook-triggered agent, a semantic search layer, a Telegram notification bot, and a scheduled report generator. All five are in my public repo — this post explains the design decisions behind each.
07. What’s Next?
Get In Touch
Although I'm not currently looking for new opportunities, my inbox is always open. Whether you have a question, a project in mind, or just want to talk about AI systems and automation — I'll try my best to get back to you.
Say hello