Ricardo Domingues

Ricardo Domingues

I like solving problems so much I've built three businesses around it.

Open to Founding Engineer Forward-Deployed AI Engineer at early-stage teams

That's the actual drive, solving something real and understanding what makes someone actually pay for it, not just shipping code. I'm looking for a team where that instinct has more to build with.

See What I've Shipped Message Me Watch Me Build
↗ July 2026 · San Francisco
Silicon Valley Fellowship 2026
~30 spots selected from 1,000+ applicants.
Location
Based in Portugal
Open to relocation — US · UAE · Europe
Companies I worked with
Legal AI, U. Coimbra spinout CuryaAI scheduling, 2,000+ workers Portugal's first electric ride-hail Ignis VaultDigital vault for legacy planning Kestra automation partner, Brazil 45 AI systems live for Portuguese SMEs FlexIQAI email lead-gen, autopilot Creative agency, Lisbon Scale Labs+€5M/year in client revenue
00

Why I'm worth a conversation

2026 · Founder

Worked with 40+ companies vibe coding their products. I pivoted to funded founders and found €100k+ in GDPR exposure across 10+ startups.

€100k+
GDPR exposure found
10+
funded startups audited
40+
companies worked with

This is the pattern I bring to a team: build something real, sell it myself, and go where the results are.

↗ SecureDev ↗ PentestAI
01

Work

tap an entry to open the case
2024 case study →
Software Developer — Pollux
Real-time radio communication for stratospheric balloon launches, up to 250 km.
250 kmreal-time radio range
2024 · Engineering

Software Developer — Pollux

Real-time radio link to stratospheric balloons
Problem

A prior project (Thestias) had unreliable long-range communication with stratospheric balloons. CAMO had to fix that and hit real-time performance.

What I did
  • Joined a 6+ person engineering team at Pollux, a student-run aerospace company, reporting to the project lead and inheriting Thestias's existing codebase and architecture
  • Diagnosed Thestias's long-range communication failures and redesigned the radio link within that existing system
  • Built CAMO on Raspberry Pi with GPS, environmental sensors, and LoRa transmission, then flew it on a real stratospheric balloon launch
Outcome
250 km
real-time radio range
Flew
on a real balloon launch
ISO/ECSS
standards compliant
Python Raspberry Pi LoRa GPS
2025 case study →
Freelance Web Developer
Booking sites for local businesses, full sales cycle, solo.
3sites delivered
2025 · Freelance

Freelance Web Development

Sell · build · deliver
Problem

Local accommodation businesses had no way to take bookings online.

What I did
  • Ran the full sales cycle solo (prospect, pitch, close, deliver) across 20+ meetings
  • Sold and delivered 3 custom websites with booking functionality and owner notifications
  • Set up a referral partnership with a local accommodation company to source warm leads
Outcome
3
sites delivered
20+
sales meetings run
€200
first sale closed

Local businesses have real purchasing intent when you solve a specific, tangible problem.

Web development Booking systems
2025 case study →
Founder — AI Automation Solutions
Built an SDR agent, ran 100+ cold outbound touches, and pitched founders at Portugal's largest fitness chain myself.
100+outbound touches
2025 · Founder

AI Automation Solutions

Built solo. Pitched solo. No sales team.
Problem

Companies had messy, manual sales and ops processes, with no one in-house to map the workflow, find the real bottleneck, and automate it.

What I did
  • Built and ran my own outreach pipeline first (lead qualification, routing, meeting booking) and used it as the live demo with every prospect
  • Ran 100+ outbound touches across cold email and calls, leading to 30+ discovery conversations with founders and CEOs
  • Mapped each business's workflow and built automations to fix the real bottleneck, starting in n8n, then moving to LangChain, LangGraph & Kestra as requirements grew
Outcome
100+
outbound touches
30+
founder & CEO conversations
3
major companies engaged

Ran conversations with Fitness Up (Portugal's largest fitness chain), CapTemp (international industrial monitoring) and Triber (brand strategy agency).

LangChain LangGraph Kestra n8n
2026 case study →
Founder — AI Security / Lusitai
Worked with 40+ companies vibe coding their products on security. Pivoted to funded founders and found real GDPR exposure.
€100k+GDPR exposure found
2026 · Founder

AI Security / Lusitai

Built for companies vibe coding. Sold to funded founders.
Problem

Vibe-coded apps (Lovable, Bolt, Cursor) were shipping with real security holes: exposed IBANs, disabled row-level security, API keys sitting in the browser. The companies vibe coding didn't see it as their problem. Founders did: a GDPR breach means fines up to €20M or 4% of revenue.

What I did
  • Built SecureDev, a security scanner SaaS (Next.js, FastAPI, Redis, PostgreSQL, 24 scanners, AI-generated fixes, Stripe pay-per-scan), and worked with 40+ companies vibe coding their products, but they didn't see security as their problem
  • Pivoted to funded startup founders instead, where a breach is existential, not theoretical
  • Rebuilt the scanner as PentestAI, a 22-phase pentest workflow on Claude Code, because the original tool was too shallow for real startup-grade audits
Outcome
10+
startups audited
€100k+
GDPR exposure found
1
paying client

Audited ByTheLaw, Curya, Tapsi & IgnisVault. Several have since received investment. Visited ByTheLaw in person; the scan caught what their existing security service had missed. One paying client was enough signal that the segment wasn't going to scale, so I moved on.

Next.js FastAPI Claude Code PostgreSQL Redis Stripe
↗ SecureDev ↗ PentestAI
July 2026 Upcoming
Silicon Valley Fellowship — SF Bay Area
~30 spots selected from 1,000+ applicants.
02

Projects

standalone builds
Architecture diagram: WhatsApp/Twilio webhook into a FastAPI API, a Postgres queue and checkpointer, and an async worker running the LangGraph agent

WhatsApp LangChain Harness

Most WhatsApp bot tutorials wire an LLM straight to a webhook and call it done — no queue, no retries, no memory between messages, no way to see what's happening. This is the infrastructure layer instead: a FastAPI webhook validates Twilio's cryptographic signature and queues incoming messages in Postgres with debounce and retry, an async worker pulls them into a LangGraph agent with per-conversation checkpointing and pgvector-backed semantic memory, and a Next.js admin panel — gated behind Better Auth and an internal service token — gives live visibility into chats, the queue, and agent state. Ships with Railway deploy and stress-testing docs. Built to be a reusable template for standing up production WhatsApp agents, not just a demo.

ricardoo022/harness_whatsapp

CrmToVoice

Real estate agents don't want to type, they want to talk. The agent tells Siri what happened after a visit or call, and the agent figures out who the client is, which property, and what happened, then writes it straight into the CRM. No forms, no app to open. Built with LangChain/LangGraph, an OpenRouter-served LLM, Airtable as the CRM, and a FastAPI webhook for Siri to talk to.

ricardoo022/CrmToVoice

RAG Pipeline

Reusable retrieval-augmented generation framework: Postgres/pgvector-backed ingestion, hybrid search, and evaluation, built to drop into new projects by swapping in new docs and eval questions. Sweeps chunking strategy, search type, HyDE, and reranker combinations to find what retrieves best, then layers a conversational agent on top for grounded, cited answers.

ricardoo022/rag-pipeline
Real test call · 00:12:23
“No, it looks great.”
85% voice
97% face
Voice said yes. The face already said no.

Sales AI Pipeline

Sales teams review calls by ear or skim a transcript, missing the hesitation, tone shift, or blank stare that actually predicts whether a deal closes. This pipeline reads a meeting recording the way a sharp sales manager would: voice, face, and words together, not just the words. Built after finding a Portuguese company (€5M+ revenue) whose sales-call analysis product only ran the transcript through an LLM; reached out unprompted, met their engineering team, and showed them the multimodal upgrade.

ricardoo022/salesAIpipeline
Lusitai dashboard showing a completed scan with a security score of 70 and severity breakdown

SecureDev (Lusitai)

Security scanner SaaS for apps built with AI coding tools (Lovable, Bolt, Cursor). Submit a URL, get an automated scan across 24 security checks, and get back a fix file written for the same AI tool to paste in and patch. Full product: Next.js frontend, FastAPI backend, Redis, Claude/GPT-4o for the fix generation.

ricardoo022/SecureDev
MoviesNow hero banner showing a featured film with rating and showtime CTAs

MoviesNow

Portuguese movie aggregator. Pulls data from IMDb and Rotten Tomatoes to show what's playing in Portuguese cinemas.

ricardoo022/MoviesNow
F1 score comparison across 12 model configurations, best result highlighted

Human Activity Recognition

ML pipeline that classifies 16 human activities from wearable sensor data (accelerometer, gyroscope, magnetometer): outlier detection, feature engineering, PCA/ReliefF dimensionality reduction, and a custom k-NN classifier benchmarked across 12 configurations with statistical hypothesis testing.

ricardoo022/Human-Activity-Recognition
Googol search engine client showing the search bar and URL indexing panels

Googol — Distributed Search Engine

Google-style search engine built for a Distributed Systems course: a gateway, multiple index "barrels," and a crawler communicate over gRPC to index pages, rank results, and track backlinks across machines.

ricardoo022/SD-PROJECT
Best configuration · 5 independent runs
Evolved weights, no gradients, no reward signal from the environment.
97% avg success rate
100/100 best individual
Genetic algorithm vs. reinforcement learning, same task.

Lunar Lander Neuroevolution

A genetic algorithm evolves the weights of a small neural network (8-12-2) to land the Gymnasium LunarLander instead of training it with gradient-based RL. Tournament selection, uniform crossover, and fitness-dependent Gaussian mutation run across parallel episode evaluations; a custom fitness function scores proximity to the pad, velocity, orientation, and leg contact. The best configuration landed successfully in 97% of test episodes across 5 independent runs, with one individual hitting 100/100.

ricardoo022/lunar-lander-neuroevolution
[ screenshot ]

PentestAI — Claude Code Security Scanner

24 Claude Code skills that turn it into an automated black-box security scanner: each phase runs as its own slash command, chaining recon through exploitation in optimized parallel waves. Powers SecureDev, the scanner built for AI-assisted app builders.

ricardoo022/PentestAI-with-claude-code
03

Skills

engineering, ML, and go-to-market — not one without the other
AI & Agents
LangChain LangGraph Claude Code RAG n8n Kestra
Engineering
Python TypeScript Node.js React Next.js FastAPI Docker PostgreSQL Redis Database
Machine Learning
scikit-learn PyTorch feature engineering PCA statistical testing audio/video ML pipelines
Go-To-Market
outbound automation lead qualification customer discovery demo building personalized proposals CRM workflows market validation pivot decisions
04

What clients say

unedited quotes, Lusitai clients

Before this site, there was Lusitai — a security scanner SaaS I built, sold, and supported myself. These are the founders and teams who used it.

“Lusitai frees you to build the first version of your app without worrying about security details. Once the features are ready, Lusitai is the perfect tool to review security and deploy with peace of mind.”

Ronnald
Ronnald
Founder · Simplework.ai

“After building multiple Vibe Coding applications, Lusitai literally increased my confidence in delivering to clients and securing their data. Before, I did manual checks, and now it's a tool that literally lives in my environment and I use it every time I do any kind of project, every week. On top of that, I talk about it to everyone I can within the community.”

Diogo Köpke
Diogo Köpke
Founder & CEO · Kopk AI

“I feel much safer even without knowing anything about security. Lusitai tells me exactly what's wrong and I just have to apply the fixes in the AI. I can develop fast, launch to production and sleep peacefully, without having to become an expert in an area that's not mine.”

Valentin Rangelov
Valentin Rangelov
Founder & CEO · FlexIQ

“Using Lusitai has significantly improved the way we approach security in our development process. Instead of treating security as a late-stage concern, Lusitai allows us to identify potential vulnerabilities early, while code is still being written. The platform fits naturally into a modern development workflow, providing clear, actionable insights without slowing teams down.”

Tapsi
Team

“Lusitai did a security check-up on our custom CRM and brought us very valuable insights. They detected flaws that could compromise client and employee data when launching the internal tool and gave clear, prioritized recommendations for fixes. Besides strengthening security, they helped us prevent compliance risks and potential fines in the future. I recommend without hesitation!”

Creative Flow
Team

“Lusitai helped us secure our MVP quickly and efficiently. The cybersecurity report perfectly complemented our knowledge, allowing us to mitigate risks. A professional, agile service with excellent results.”

M
Mafalda
Founder · Ignis Vault

“Security in a project is always a priority and Lusitai's solution fit like a glove. The analysis was fast and the fixes were on point, clear and simple to implement.”

Roberto
Roberto
Vibe Coder
05 · let's build

Open to relocation, anywhere.

If you're building something early and need someone who can ship it, sell it, and stay close to customers — let's talk.

ricardodomingues.166@gmail.com LinkedIn GitHub YouTube
© 2026 Ricardo Domingues — built & deployed solo.