AI agents executing
Autonomous agents running tasks in real time: each takes a role and collaborates on the flow.

No marketing spin: a real look at what I build — neural networks, autonomous agent ecosystems, and the environment where architecture is born.
Real screens: Nova OS (orchestration, RAG, cost, health), MWS operational CRM (SECOP/scraper funnel, lead detail, finance) and MWS AI (WordPress/WooCommerce sales agent).
Open full sizeFunnel view with KPIs (active, estimated budget, SECOP, scraper/AI), filters by status/temperature/source/territory and an operational leads table.
Open full sizeLead record with temperature, probability, source, territory and requested service; actions to return to the funnel or convert to client.
Open full sizeIncome, expense, balance and margin dashboard in COP with prior-period comparison and daily cash flow. The same admin opens the CRM funnel.
Open full sizeProduct landing: AI Sales Assistant widget with RAG over live inventory. Answers stock/sizes, shows a product card, and runs 24/7 with a <2s response target.
Open full sizeChannel for agencies/distributors (volume licenses) and a bridge to custom AI agents from Medellín Web Soluciones when scope goes beyond the WordPress plugin.
Open full sizeEmbeddable chat shell from the plugin: brand header, message thread and agent bubbles ready to connect to the MWS AI SaaS.
Open full sizeNova agency graph: 5 areas, 29 roles and 35 nodes. Real-time execution timeline on the left, crew directory on the right, where every node is a specialist with its own configuration.
Open full size13 system topologies documented inside the product itself and versioned in git: runtime, integrations, data and UI. Architecture is read where the system is operated, not in a separate PDF.
Open full size12.2M tokens and 296 crew calls at an estimated USD 29.90. A 14-provider cascade mixing free tiers and local Ollama, with share and trend per model to decide where each task runs.
Open full size2,749 chunks indexed in the nova_knowledge collection, 6 domain packs and 159 first-party documents. RAG validation, per-category breakdown and knowledge map before moving to fine-tuning.
Open full sizeEleven monitored subsystems with status and active model: LLM, STT/TTS, vision, YOLO, toolkits and automations. The capture shows 10/11 online and one down — the panel exists to surface that, not to hide it.
Open full size202 agent profiles from MIT-licensed open source catalogs, grouped into 17 areas and mapped onto Nova's role model. Browsable as a graph to decide which specialist handles each task.
Visualization of neural network behavior: nodes, connections and propagation that power production models.
Autonomous agents running tasks in real time: each takes a role and collaborates on the flow.
Topology of an agent network: communication, work split and convergence to a useful result.
Nova's operational architecture: CEO, divisions and 29 specialists as a system—not a demo.
LEXIA operational surface: API + OS + analytics for AI-assisted legal work.
Operator panel for opportunity radar, scoring and approval before Woo publish.
The real environment where we design, build and test: the workshop behind every solution.
The build process condensed: from the first line to a running system.