LinkedIn | Portfolio | Email Me

Python is my peak weapon. I engineer highly-resilient, production-ready AI agents, MLOps platforms, and custom software infrastructure. Based in Badajoz, Spain.

🌐 Official Website & Interactive Showcase: wisrovi.dev

In modern engineering, the bottleneck is rarely writing code—it is resiliency, scalability, and design integrity.

While many build simple AI wrappers, my primary focus is designing high-availability systems where Generative AI, Computer Vision, and modular infrastructure converge to solve complex enterprise problems.

By leveraging Python as my supreme engineering backbone, I create custom frameworks, distributed MLOps clusters, and multi-agent state machines designed to survive in high-concurrency production workloads.

I design next-generation cognitive systems that move past simple prompt engineering into complex, stateful reasoning loops.

• Stateful Agent Orchestration: Building multi-agent systems using LangGraph and LangChain to orchestrate structured, reliable agent behaviors.
• RAG (Retrieval-Augmented Generation): Engineering advanced semantic search and query expansion pipelines, utilizing vector embeddings and vector databases (Milvus, PgVector) to inject contextual accuracy.
• Generative AI & LLMs: Implementing LLM integration patterns, reasoning validation, fine-tuning structures, and safety/security filters.

I believe in writing code that other developers rely on. To solve common bottlenecks in database operations, caching, and pipeline orchestration, I built a catalog of 26+ custom Python libraries:

• Orchestration (wpipe): A lightweight Python pipeline executor featuring WAL-mode SQLite state storage, GIL bypass, DAG scheduling, and dynamic checkpoints.
• ORM Systems (wsqlite, wredis, wclickhouse): Custom wrappers resolving database concurrency, soft deletions, and bulk insertions utilizing Pydantic v2 schemas.
• Zero Trust Security (wauth, wFabricSecurity): Cryptographic vaults tying encryption to salted machine hardware signatures and securing distributed networks.

Scaling training campaigns from a single workstation to distributed clusters.

• MLOps Automation: Complete training lifecycle management of YOLO and RT-DETR models.
• Distributed Compute: Celery-based worker structures mapping and utilizing GPU topologies, autobatching weights based on VRAM capacity, and tracking runs via MLflow.

NeuralForge AI (train_service2) is the ultimate demonstration of my engineering philosophy. It is a distributed hyperparameter optimization and training ecosystem built entirely on top of the wisrovi SUITE, now updated to v2.0.0.

The system utilizes a decoupled microservices architecture designed to scale training campaigns from a single workstation to multiple GPU-accelerated Docker nodes:

• FastAPI & Gradio Gateway (wyoloservice2_control_server): Validates configurations, injects path metadata, and manages study lifecycle through an intuitive dashboard.
• Optuna Study Manager (wyoloservice2_manager): Orchestrates distributed hyperparameter searches, query suggestions from a central PostgreSQL instance, and evaluates trials.
• Celery-Docker Invoker (wyoloservice2_invoker): A worker daemon listening on priority queues that creates ephemeral directories, writes trial configurations, and spawns containerized training tasks.
• Ephemeral Training Containers (wyoloservice2_worker): Dockerized runtime executors that pull raw configs, download S3/MinIO datasets, execute YOLO training loops, report real-time telemetry to MLflow, and automatically clean up resources on exit.
• React Dashboard (NeuralForgeAI): Interactive and rich user interface built with React to easily launch, track, and analyze training studies.
• Strict Redis Priority Queues: Workloads are dynamically routed using a prioritized queue system: private_queue (worker_*) > gpus_high > gpus_medium > gpus_low ensuring immediate response for targeted hardware debugging.

Component	Category	Registry / Status	Focus & Key Technical Specs
wpipe	Core Engine		WAL SQLite state storage, GIL bypass, thread/process parallel setups.
wpipe-mcp	Generative AI		Model Context Protocol (MCP) server letting LLMs program and deploy pipelines.
wsqlite	Database ORM		TableSync migration generator, soft deletion query filters, connection pooling.
wauth	Security		Secrets vault tied deterministically to machine hardware signatures (Fernet).
wredis	Database ORM		Sync/async Redis integration, cache expiration decorators, rate-limiters.
wFabricSecurity	Security		Zero Trust security system for Hyperledger Fabric (ECDSA signatures).
wyolo	MLOps		MLflow + S3 automation wrapper for YOLO & RT-DETR models.
ProcessAudio	MLOps / Audio		Librosa feature extraction & audio data augmentation.
wticket	Support / Helpdesk		Helpdesk manager with dynamic SLAs and client portal APIs. Live demo: wticket.wisrovi.dev.

Explore all 26+ packages in my complete library repository.

• Wisrovi Legacy: A procedural 3D driving RPG built with Vanilla JS and WebGL/Three.js. Demonstrates my frontend abilities and procedural texture rendering.

A comprehensive 22-page technical guide has been compiled and is available as a PDF in this repository:

• Location: main.pdf
• LaTeX Source: docs/main.tex
• Contents: Detailed architecture diagrams of the w-libraries suite, step-by-step installation routines for docker/GPU runtimes, Optuna-Celery configurations, and future roadmaps with LangGraph/RAG integrations.

If you find my open-source projects or guides helpful, feel free to support my ongoing research:

"I love to research and learn as much as I love chess or my family."