VazquezDev
/

RAGAGENTSAUTOMATION

AI Systems Architect & Developer_

I engineer production-grade RAG architectures and multi-agent workflows on private infrastructure.

Secure, scalable systems built to automate high-stakes operations while keeping sensitive data on company-controlled servers.

VAZQUEZDEV.PRO
ONLINE
Adrian Vazquez Vazquez - AI Systems Architect & Developer.
PROFILE

#About Me_

VazquezDev OS v2.4.0-pro
Online
BOOT_LOG
[ OK ]Initializing VazquezDev OS v2.4.0-pro...
[ OK ]Loading AI_Architect_Kernel...
[ OK ]Establishing secure bridge to Barcelona-HQ...
[ OK ]Mounting /mnt/knowledge/private_rag...
[ OK ]System status: OPTIMAL
KERNELAI_ARCHITECT
LOCATIONBCN/ES
UPTIME∞ RUNNING
FOCUSPRIVATE AI
010203040506070809101112131415161718
adri@vazquezdev:~$whoami
Adrian Vazquez//AI Architect & Developer
adri@vazquezdev:~$cat mission.md
I build AI that runs on your servers, not someone else's cloud. RAG pipelines, autonomous agents, and orchestration workflows — shipped to production, not stuck in demo mode.
adri@vazquezdev:~$./capabilities.sh --privacy-first
PRIVACY[OK] Privacy by Design — on-premise and company-controlled deployments
PROD[OK] Production First — robust architectures beyond prototypes
E2E[OK] End-to-End Ownership — vector layer, backend, orchestration and UI
TEAM[OK] Team & Delivery — integration with engineering teams, agile methodologies, and CI/CD
#For an interactive deep dive into my project history and stack, ask Cortex AI. brain.vazquezdev.pro
adri@vazquezdev:~$

>> // SYSTEMS DEPLOYED _

Production systems and technical case studies. Built for real users, real constraints, and measurable operational impact.

[RAG-01]
Live · Deployed
VazquezDev·RAG Architecture · Vector Search · FastAPI
01

Cortex AI

An interactive RAG system integrated directly into the portfolio that allows recruiters and technical leads to audit my experience, architecture decisions, and stack in natural language.

01SITUATIONThe Problem

The Problem:

A traditional portfolio is a passive document. Recruiters and potential clients spend seconds scanning it — and leave without grasping the depth of the work. VazquezDev needed a way to turn passive visitors into active conversations: answering specific questions on demand without forcing anyone to read page by page.

  • Static pages can't answer "what's your RAG stack?" or "have you done this before?"
  • VazquezDev's work spans too many dimensions to fit in a simple scroll
  • Most portfolio visitors leave with an incomplete picture
02OBJECTIVEThe Objective

Objective:

Turn vazquezdev.pro into an interactive layer where anyone can query the full professional profile — experience, projects, stack, and services — in a single natural-language flow:

  • One entry point for everything: experience, projects, stack, services
  • Multilingual from day one
  • Fast, low-latency answers — no waiting
03SOLUTIONThe Tech

Stack & highlights:

  • Frontend: Vue 3 · Vite
  • Backend: FastAPI · Python
  • Vectors: ChromaDB — lightweight self-hosted storage
  • LLM: Groq API — ultra-low latency, near-zero inference cost
  • Ops: Docker · Traefik · VPS
  • Fully decoupled frontend and backend
  • Knowledge base built from professional history and real projects
  • Conversational access to the full professional context behind this portfolio
04RESULTSThe Impact

Outcome:

vazquezdev.pro now behaves like a conversational assistant over curated professional context:

  • Self-hosted vector layer — portfolio knowledge stays server-side
  • Ultra-low latency inference through Groq
  • Each layer of the stack evolves independently
[AGT-01]
Live in Production
CSR-Online·Agentic Workflow · Voice-to-Text · Orchestration
02

Telegram Reporting Agent

An asynchronous orchestration pipeline that ingests unstructured voice notes via Telegram, classifies intent, and generates technical PDF reports, eliminating hours of weekly administrative overhead.

01SITUATIONThe Problem

The Problem:

After every client visit, sales reps spent 1–2 hours writing PDF reports by hand — formatting, attaching files, sending emails. The information was fresh right after the meeting; the bottleneck was transcription and bureaucracy. Selling time was being consumed by word processing.

  • 1–2 hours lost per report, per rep, after every single visit
  • Inconsistent formats depending on who wrote the report
  • Reports sent late — or skipped entirely — when the day got busy
02OBJECTIVEThe Objective

Objective:

Eliminate manual report writing entirely. A voice note or free-form text should be enough — the system produces a consistent, professional PDF and delivers it automatically:

  • Auto-detect five report types via Claude prompt design
  • Server archive plus automatic email delivery
  • Full chain: transcribe → classify → HTML → PDF → email → confirm
03SOLUTIONThe Tech

Stack & highlights:

  • Orchestration: n8n
  • Speech: Groq Whisper large-v3-turbo
  • Generation: Claude Sonnet 4.6 (HTML report)
  • PDF: Gotenberg
  • Mail: SMTP relay · Exchange on-premise
  • Infra: Docker · VPS
  • Five report types auto-classified from messy voice notes
04RESULTSThe Impact

Outcome:

A production workflow that pays back from the very first report:

  • Before: 1–2 hours lost per rep, per visit
  • After: one voice note. The system handles the rest
  • 6 active users in production at CSR-Online
  • Zero training required for adoption
[WRK-01]
Live in Production
Personal·Autonomous Agent · NLP Extraction · ETL
03

Invoice PDF Bot

A multi-state autonomous agent that extracts unstructured natural language data, issues PDF invoices, and syncs Drive and Sheets APIs, collapsing the billing cycle to a single interaction.

01SITUATIONThe Problem

The Problem:

Before automation, a typical freelance billing loop meant 20+ minutes per invoice: open accounting software, fill out forms, export the PDF, save to Drive, log it in a spreadsheet. A process you can describe in half a minute still required five tools and constant context-switching. At a few invoices a week, that overhead quietly eats a full workday every month.

  • 20+ minutes per invoice in repetitive, zero-value admin
  • Jumping between accounting UI, Drive, and Sheets on every single issue
  • Mental overhead of remembering to log, save, and file every invoice correctly
02OBJECTIVEThe Objective

Objective:

Let a solo freelancer issue compliant PDF invoices directly from their phone, with near-zero friction — one message in, one PDF out, everything filed automatically:

  • Parse messy natural language reliably every time
  • Deliver the PDF instantly back in Telegram
  • Keep Drive and Sheets automatically aligned for taxes and history
03SOLUTIONThe Tech

Stack & highlights:

  • Orchestration: n8n
  • Extraction & copy: Claude API
  • Channel: Telegram Bot API
  • Storage: Google Drive API · Google Sheets API
  • Infra: Docker · VPS
  • Full cycle: extract → PDF → Telegram → Drive → Sheets
04RESULTSThe Impact

Outcome:

Practical automation built for daily use — not a demo:

  • 3–4 invoices per week in steady production use
  • Every invoice traceable in Sheets and archived in Drive
  • Full billing loop handled from one Telegram message
  • 20 minutes → 30 seconds, every single time

>> // CAREER PATH _

From full-stack engineering foundations to private AI systems in production.

Full Stack Eng.2023
Backend & Data2024
AI Architect2025
Head of AINOW
Mar 2026PRESENT
01
CURRENT FOCUS

AI Engineer

CSR Online·Mar 2026 - Present
  • Leading the company's AI department and reporting directly to the co-founder. Responsible for defining the AI strategy and executing it end-to-end.
  • Design and implementation of agentic workflows and enterprise RAG architectures on-premise.
  • Full lifecycle ownership: infrastructure, vector database, backend, and production deployment.
Oct 2025PRESENT
02
ACTIVE

AI Systems Architect & Developer

Vazquez Labs·Oct 2025 - Present

AI agency focused on bringing enterprise-grade systems to companies that can't afford to experiment.

Vazquez Labs designs and deploys autonomous agents, RAG pipelines, and workflow automation systems — built on private infrastructure, shipped to production, and engineered to last.

Independent agency, operating alongside my role as AI Engineer at CSR Online.

Sep 2023Jan 2026
03
COMPLETED

Full Stack Product Engineer

VazquezDev·Sep 2023 - Sep 2025

The stage where I went from writing code to engineering complete digital products — and where VazquezDev was born.

End-to-end design and delivery of 6 B2B products with no handoffs: from architecture and database to frontend, deployment, and production.

The foundation that now lets me build AI systems from end to end.

Sep 2024Jul 2025
04
COMPLETED

Backend & Data Infrastructure Engineer

Senyum SL·Sep 2024 - Jul 2025
  • Databases (SQL Server): administration; advanced T-SQL (queries, views, indexes, stored procedures); performance tuning.
  • C#/.NET (a3ERP): data automations and integrations.
  • Power BI: data modeling and dashboard reporting.
  • Distributed Systems: Backend infrastructure maintenance for high availability.
STACK INITIALIZED

// TECHNOLOGY STACK

The stack I use to ship private AI systems in production. No hype, no filler.

LAY-0106 TOOLS
The Engine

Infrastructure & Backend

Where business logic, data persistence and deployment live.

FastAPIHigh-speed Python APIs for AI systems and internal tools
DockerContainerized deployment for stability
PostgreSQLRelational databases for production workloads
RedisSemantic caching and queue management
TraefikEdge router and reverse proxy for microservices
VPS (Servers)Linux server operations, networking, SSL, deployments and monitoring
LAY-0206 TOOLS
The Brain

AI & Orchestration

The reasoning and workflow layer behind RAG systems, agents, and automations.

PythonThe native language of AI infrastructure
LangGraph / LangChainGraph architecture for agents with memory, loops and conditional logic
ChromaDB / QdrantSelf-hosted vector memory for RAG and semantic retrieval
n8n · MCP ProtocolWorkflow orchestration and native connection of external and internal tools
LLMs — Claude · Groq · GeminiStrategic selection based on latency, cost, and required privacy level
Ollama · Hugging FaceLocal LLM inference — zero third-party dependency, data stays in your infra
Architecture principle: Private infra → Vector layer → LLM reasoning → Human approval
LAY-0304 TOOLS
The Interface

Human-Agent Interface

Modern interfaces for humans to supervise and control AI.

JavaScript / TypeScriptType-safe frontend development
Vue 3 / VuetifyEnterprise UI with SSR optimization and SEO-ready architecture
NuxtSSR, routing, metadata and production-ready Vue applications
Tailwind CSSClean and responsive interface design
> DATA_FLOW:[ INFRASTRUCTURE ][ AI CORE ][ INTERFACE ]_
SYSTEMS ACTIVE

// GLOBAL REACH

Built to operate across international teams, markets and time zones.

SYS-0103 LANGUAGES
LANGUAGES
ES
EspañolNative
CAT
CatalàNative
EN
EnglishFull Professional
SYS-02 ONLINE
BASE OF OPERATIONS
COORD:41.3874° N · 2.1686° E
Barcelona, Catalonia, Spain

One of Europe's top tech & startup hubs. Direct access to the Mediterranean innovation ecosystem.

GMT+1 Remote-first Available globally