Español

Sergio Márquez

Backend & AI Engineer · Agents, RAG and LLMs · Badajoz, Spain

I build AI agents and systems that make it to production.

About

At VITALY I design and run LLM systems on Google Cloud, from multi-agent assistants with RAG and voice to document validation and data pipelines. I come from backend engineering and treat every model-backed system as one more service: designed, evaluated against data and operated with observability. I have coded with agents every day since 2024.

Projects

Publishing and generation

  • One dAIly Blog

    Autonomous blog: an AI pipeline researches, writes, illustrates and publishes a technical article every day, with no manual review.

    Claude · Codex · fal.ai FLUX · Astro · Node.js · PostgreSQL

    blog.sergiomarquez.dev

  • Short AI videos

    A daily series of vertical videos explaining AI concepts, produced by my own pipeline: script, voice, animation and captions.

    TypeScript · Remotion · ElevenLabs

  • AI Video Ad Generator

    From a text brief to a complete 9:16 vertical video ad: Gemini script, ElevenLabs voice, CLIP-reranked b-roll, and ffmpeg render. Fully automated.

    Gemini 2.5 Flash · ElevenLabs · fal.ai FLUX · Replicate CLIP · ffmpeg · Next.js

  • AI Photo Transformer

    AI portrait transformation: Gemini classifies and enhances the prompt and Seedream 4 generates the 4K image, with streaming progress (SSE).

    FastAPI · Python · Gemini 2.0 Flash · Replicate · Seedream 4 · PostgreSQL

Conversational agents

  • WhatsApp AI Agent

    Self-hosted WhatsApp agent: hot-swappable LLM, Whisper voice-note transcription, at-least-once delivery, and multi-arch Docker deployment.

    TypeScript · Baileys · Next.js · Whisper · SQLite · Docker

  • AI Voice Receptionist 24/7

    Conversational phone agent (ASR + LLM + TTS loop) that answers calls, handles questions and manages appointments by voice, integrated with telephony and scheduling.

    ElevenLabs Conversational AI · Gemini · Cloudflare Workers · SIP · GoHighLevel

  • Personal AI Agent (Google ADK)

    Open source

    Multi-agent system with Google ADK (orchestrator and specialists) that answers about my CV and searches my blog with real tool use.

    Python · FastAPI · Google ADK · Gemini · Docker

Tools

  • acestream-docker-home

    Open source

    One-click Acestream on Windows, Linux and macOS via Docker, with a bilingual (EN/ES) web player.

    Docker · Python · Shell

  • Multimodal Vector Search

    Open source

    Semantic search over text, image, PDF with OCR, audio, and video using gemini-embedding-2 on Vertex AI and MongoDB Atlas.

    Python · FastAPI · Vertex AI · gemini-embedding-2 · MongoDB Atlas

  • Video Transcriber & AI Summarizer

    Open source

    CLI for CUDA-accelerated transcription with Whisper, multi-provider AI summaries, and LinkedIn/Twitter content generation.

    Python · Whisper · Gemini · PyTorch · CUDA

Games

  • One Bad Wire

    Puzzle game: fix the wiring of a telephone exchange before everything fails. Built with Godot, with an online leaderboard and a web build.

    Godot · GDScript · Claude Code

  • Pingufly

    Launch a penguin as far as you can. A tribute to Yeti Sports, rebuilt in 3D with Godot and AI-generated models.

    Godot · Meshy · Blender

More projects on GitHub

Experience

VITALY

AI Engineer

May 2025 – Present

Applied AI end to end: assistants, documents, data and operations.

  • End-to-end owner of a multi-tenant conversational assistant in production, with a multi-agent architecture (Google ADK + LiteLLM), RAG, multi-provider model routing and voice (STT/TTS).
  • Automated RAG quality evaluation (LLM-as-judge) with dataset generation at ingestion.
  • GCP data platform (BigQuery, Dataform, Dataflow) with cost-controlled backfills and alerting, plus the API that serves that context, encrypted, to the assistants.
  • GitOps on multi-cluster GKE, GitLab CI/CD and observability with OpenTelemetry and New Relic.
More about this role
  • Document validation with multimodal models and ongoing provider comparison (Gemini, Claude and GPT).
  • Agents that turn a phone call into structured data (Whisper + Gemini) and hybrid (rules + LLM) PDF extraction.
  • Internal app to govern AI use cases, with LLM pre-assessment, from architecture to demos with the business.
  • Business automations in production and quality standards for the team's repositories (tests, SAST, CVEs and CI).
  • AI-assisted development with Claude Code (agents, skills) and multi-model review before critical changes.

Stack Python · FastAPI · Google ADK · LiteLLM · Vertex AI · BigQuery · Dataform · Dataflow · MongoDB · PostgreSQL · Firestore · Oracle · GKE · GitLab CI/CD

Full-Stack Developer

Jun 2021 – May 2025

Java/Spring backend leading GKE migration and CI/CD.

Role details
  • Strategic Middleware APIs: Designed and developed Spring-based middleware to optimize frontend/external services communication.
  • GKE Migration & Optimization: Led the migration to Google Kubernetes Engine, automating continuous delivery and optimizing deployments.
  • Post-Merger Data Integration: Actively participated in the data integration after the Preving/Cualtis merger, performing mapping and application adaptation.
  • International Experience (Dublin, 2022): Spent 3 months working remotely from Ireland, improving English fluency and adaptability in global environments.

Stack Java · Spring Framework · Angular · Docker · Kubernetes · GCP · GitLab CI/CD

Intern Developer

Apr 2021 – Jun 2021

Intensive training and first contact with professional web development environments.

Role details
  • Received intensive training in Java Spring, Angular, Docker, and Kubernetes.
  • Participated in agile development teams, contributing to real-world projects and learning full software lifecycles.

Certifications & training