Privacy-First AI Infrastructure

Meet
JuanaIA

A fully self-hosted personal AI assistant — no cloud, no API keys, no data leakage. Running entirely on local hardware with production-grade architecture.

Explore the project →
scroll to discover

The Project

What is JuanaIA?

JuanaIA is a fully self-hosted personal AI assistant built on a microservices architecture with local LLM inference, semantic memory, RAG pipelines, and an autonomous agent loop. No external dependencies — everything runs on an NVIDIA RTX 5090.

Inference

Qwen3-Omni 30B

vLLM · AWQ 4-bit · 134K ctx
NVIDIA RTX 5090 · 28GB VRAM

Memory

Semantic + Episodic

pgvector · BGE-M3 embeddings
Hybrid search · ParadeDB

Agent Loop

Autonomous Planner

LangGraph · Kafka backbone
Replanning · Tool execution

Interfaces

Multi-Channel

REST · SSE streaming · PWA
Voice: Whisper STT + XTTS-v2

Infrastructure

Self-Hosted Stack

Spring Cloud · Keycloak · n8n
OpenBao · Grafana · Docker

Roadmap

Coming in R4–R5

FLUX.1 image gen · Neo4j graph
Avatar (VRM) · TensorRT-LLM

Python Java / Spring Boot React 19 TypeScript PostgreSQL + pgvector Kafka Redis Docker Cloudflare Tunnel WSL2 ONNX INT8 safetensors

The Builder

Katherine E. Aguirre

Senior Software Engineer and AI Infrastructure Architect based in Buenos Aires, Argentina. Sole developer, architect, and PM of JuanaIA — designing and building production-grade AI systems end to end.

Open to remote Senior Software Engineer, AI Infrastructure Engineer, or Backend Architect roles.