João Víctor López
AI engineer and founder from Brazil, building production systems at the intersection of domain expertise and deployable intelligence. Stanford visiting student, YC Startup School alum, and the builder behind Speco — a post-training control plane shipping at specoai.com.
Currently building
Speco: the post-training control plane.
Turn domain knowledge into deployed specialist AI systems through a single platform.
Currently building
Speco
A post-training control plane that turns proprietary domain knowledge into deployed specialist AI systems.
Professional role
AI Integration Engineer
Shipping agentic workflows, retrieval systems, and operational AI inside FI Group.
Formative chapter
Stanford Summer 2024
A formative bridge between Brazil, Silicon Valley, systems thinking, and founder ambition.
Engineering depth
End-to-end AI delivery
Production systems across computer vision, retrieval, fine-tuning, and product-grade deployment.
Story
About
João Víctor López Matias is an AI engineer and founder from Ceará, Brazil, building at the intersection of systems, product, and interface. He studies Computer Engineering at Unifor, has shipped production AI systems across multiple domains, and approaches software as something that should feel both intelligent and intentional.
In 2024, he became a visiting student at Stanford, where he studied high-performance computing, energy, creativity, and startup thinking inside the environment that shaped Silicon Valley. The experience deepened his conviction that technology should not be treated as isolated code, but as a vehicle for products that change how people think, work, and create.
Before and around that chapter, João built a reputation in Brazil for taking initiative early. He was selected as a Huawei ICT Academy ambassador, founded the Huawei Club Unifor, earned national recognition through the program, and used those opportunities to align education, community-building, and emerging technology around real outcomes.
Professionally, his work spans computer vision, retrieval systems, legal AI research, and agentic internal tooling. In earlier production AI roles, he shipped computer-vision systems; in research, he worked on legal classification and hybrid retrieval; and at FI Group he focuses on internal AI workflows, RAG infrastructure, and deployment-minded systems for real business teams.
Today, that technical depth is converging with a stronger founder identity. Speco represents that next phase: a post-training control plane that turns proprietary domain knowledge into deployed specialist AI systems. Upload data, build datasets and evals, benchmark quality, and deploy through API and MCP — all in one platform at specoai.com. The throughline across all of João's work is the same: build products that are elegant on the surface, rigorous underneath, and meaningful enough to matter in the real world.
Stanford, summer 2024.
High-performance computing and systems exposure during the Stanford experience.
Press and public proof
Public signals around João's trajectory.
article
Unifor spotlight on João's Stanford exchange
Unifor documents João as a Computer Engineering student completing a summer exchange at Stanford in 2024, highlighting his AI focus, Huawei background, and immersion in Silicon Valley entrepreneurship.
article
Unifor coverage of Huawei ICT Academy recognition
Unifor reports that João was selected as one of six Huawei ICT Academy ambassadors in Brazil and later received standout ambassador recognition, tied to founding the Huawei Club Unifor and expanding technology education initiatives.
article
Unifor interview on João's international technology path
A later Unifor interview frames João's Stanford experience, applied AI work, Huawei recognition, and FUNCAP legal AI research as part of a broader international trajectory in technology.
social
LinkedIn profile
Public professional profile describing João as an AI engineer focused on scalable RAG, agentic AI, and computer vision, with Stanford and EuroTech listed among his signals.
Turn domain knowledge into deployed specialists.
Speco is what I'm building right now — a post-training control plane. Upload proprietary data, build datasets and evals, choose the right optimization path, benchmark quality, and deploy through API and MCP.
Foundation models are powerful but generic. Speco gives you the control plane to make them exceptional at your domain.
How It Works
Six steps from raw data to deployed specialist.
A complete pipeline that handles ingestion, dataset creation, evaluation, strategy selection, optimization, and deployment — in one platform.
Ingest domain data
Parse, clean, and index proprietary files. PDF, docs, structured data — all grounded to your domain.
Build dataset
Structured, versioned, inspectable training data. Every example traceable to its source.
Generate evals
Auto-generated eval suites with domain-specific rubrics. Quality is measured before deployment.
Recommend strategy
Speco analyzes your data and recommends whether prompt engineering, RAG, or fine-tuning is the right path.
Optimize & benchmark
Run optimization, measure pass rates, latency, and cost. Per-case scoring breakdowns across iterations.
Deploy & serve
One click from optimized model to live endpoint. API keys, usage tracking, and MCP integration included.
Product Demo
See Speco in action
From the public product experience to the operational dashboard. Explore every surface of the platform.
The public-facing product experience: hero, workflow, benchmarking, MCP integration, and pricing.
Hero & Value Proposition
Turn domain knowledge into deployed specialists. Upload data, build datasets and evals, choose the right optimization path, benchmark quality, and deploy through API and MCP.
Pipeline Overview
A real specialist pipeline: 94.2% pass rate, 0.87 mean score, 340ms latency, and 12.4k requests served. Live API endpoint ready for inference.
Six Steps to Deployed Specialist
From raw domain data to a deployed healthcare policy specialist: ingest, build dataset, generate evals, recommend strategy, optimize, and deploy and serve.
Why Speco
The missing layer between foundation models and production. Training abstraction, dataset control, eval-first workflow, strategy recommendation, deployment-ready, and MCP-native.
Control Plane Overview
Everything you need in one control plane: specialist management, pipeline runs, deployment endpoints, and MCP integration for agent-ready specialists.
Benchmarking & Quality
Measurable, not magical. Auto-generated eval suites, per-case pass/fail breakdowns, latency tracking, and benchmark trends across optimization iterations.
MCP Integration
Every deployed specialist is automatically exposed through Model Context Protocol. Downstream AI agents can discover and use specialists as tools, resources, and prompts.
Pricing Plans
Simple, predictable pricing. Free tier for exploration, Growth at $49/month for production teams, and custom Enterprise plans for scale and compliance.
Bottom CTA
Build specialized agents from your own data. From raw files to deployable AI systems in one control plane.
Why I Built This
The missing layer between foundation models and production.
Foundation models are powerful but generic. Every production deployment needs domain-specific quality, measurable accuracy, and operational infrastructure. Speco gives you the complete control plane to make AI exceptional at your domain.
Auto-generated eval suites with domain-specific rubrics
Per-case pass/fail with detailed scoring breakdowns
Latency and cost tracking per deployment
MCP-native integration for agent ecosystems
Training abstraction
Define your specialist, upload data, and Speco handles the rest. No infrastructure to manage.
Dataset control
Structured, versioned, inspectable training data. Every example traceable to its source.
Eval-first workflow
Quality is measured before deployment. Auto-generated eval suites ensure domain requirements are met.
Strategy recommendation
Speco analyzes your data and recommends whether prompt engineering, RAG, or fine-tuning is the right path.
Deployment-ready
One click from optimized model to live endpoint. API keys, usage tracking, and health monitoring included.
MCP-native
Deployed specialists are automatically available as MCP tools for downstream AI agents.
Selected Work
The products and systems behind the trajectory.
Collections
A founder archive organized by chapter.
Stanford, YC, Paris, infrastructure, and product-building moments grouped into intentional collections.
Connect
Let's build something ambitious together.
Open to founder conversations, product collaborations, and ambitious AI work. Explore the archive, try Speco, or reach out directly.
Late-Night Build Log