Web, mobile and AI
From the first draft to live.
I'm Júlio Machado, a software engineer for over seven years. I live in Portugal and work with companies anywhere.
EXPERIENCE AT ORACLE, GOV.PT / AMA AND GOLDEN WEALTH MANAGEMENT
Skip the animation ↓The path
How I got here.
The path
The start
Curious and self-taught from an early age.
By around 14 I was already into cybersecurity, blockchain, Kali Linux and scripts.
Learn more
Origin. It started with curiosity. At 13 or 14 I was already testing security on the family computer and had installed Kali Linux. That's how I found blockchain: I mined Bitcoin, Ethereum, Dogecoin and Litecoin, and spent hours in cryptography and security communities.
Before my first job. By the time I got my first job I had around 70 repositories: mobile apps, front end, back end, study projects and a lot of experiments.
Also. 2nd place, as a pair, in my university's programming olympiad, against campuses from all over Brazil.
- STACK
- C/C++JavaScriptNode.jsReactReact Native
- RESULT
- Today, over 180 repositories and 5 organisations on GitHub
Flow Podcast · Brazil
I gave their digital team a hand.
The path
Oracle
My first contract at 19, as a consultant for Oracle, outside my home country.
Two projects: the front end of a huge back office, then the back end in Java.
Learn more
- DONE
- The front end of a large-scale back office, then REST APIs integrated with machine learning, with a team across three countries
- STACK
- JavaSpring BootSpring Data JPAREST APIsMavenJavaScriptOracle JETKnockoutJS (MVVM)JRAF (Oracle Retail framework on Oracle JET)
- RESULT
- A back office for two large companies in Europe
On my own · 2021 to 2022
I built sites, apps and back offices.
For a while, I worked closely with the founder of Agência Digital x3 on projects for its clients.
Learn more
Content and data. Prismic CMS and Firebase (Firestore and Storage): data models, security policies and publishing workflows.
Authentication. OAuth with Google and Facebook, access control and session management.
Back office. Admin dashboards for clients: inventory, orders and reporting.
Performance and SEO. SSR/SSG in Next.js, lazy loading, code splitting, optimised images and technical SEO (metadata, sitemaps, routing).
Marketing. Meta Pixel, conversion events, UTMs, landing pages and A/B tests for acquisition.
- STACK
- ReactNext.jsNode.jsPrismicFirebase
- RESULT
- Websites, back offices and acquisition funnels live for the agency's clients, from design to operations
The path
Gov.pt / AMA
I built the front end of the gov.pt design system components, with a focus on accessibility.
I also worked on Balcão do Condutor, the online service for drivers.
Learn more
- DONE
- React components for the design system, many of them documented in Storybook
- STACK
- ReactTypeScriptStorybookWCAG 2.1 AA
- RESULT
- Used by several citizen services, to WCAG 2.1 AA
The path
Golden Group
I helped build the operation's back office at Golden SGF and connect it to external providers.
A lot of manual work started running itself. Today I'm still there in dev and project analysis.
Learn more
- DONE
- The operation's back office at Golden SGF (Golden Group, over €2B under management), provider integrations, process automation, data, feasibility analysis and internal project consulting
- STACK
- .NET / C#ReactTypeScriptAzureAzure FunctionsSQLREST APIsJWTCI/CDAutomated tests
- AI
- AI agentsMCP and tool callingAI automation
The path
My own products and client work
I have a few products of my own with partners.
When a project calls for it, I lead and guide teams across different areas.
Learn more
- DONE
- A few products with partners, web, mobile and AI projects for companies, and my own tools to test, automate and improve how I work. Everything in Build goes into them
- STACK
- Next.jsReact NativeExpoNode.jsTypeScriptPostgreSQL
- AI
- Multi-agent orchestrationRAGMCP and tool callingStructured outputsEvals and LLM-as-judgeSeveral models: Claude, GPT, Gemini and local ones
Understand
First, understand.
Before coding, I understand the problem and write a short plan.
Learn more
What problem does this solve, and for whom?
What happens if we do nothing?
What can't be missing from day one?
How will we know it worked?
Beyond code
Outside of code, I study people and business. It helps me understand what the client needs before I write any code.
Learn more
People. Human behaviour, communication and storytelling, ethical persuasion, body language, emotional intelligence.
Business. Business models, validating ideas with real customers, lean startups, strategy and positioning, growth and scale.
Sales and negotiation. Consultative selling (asking the right questions), negotiation, understanding what the client really needs.
Decisions and money. Financial literacy, risk and uncertainty, habits and discipline.
Foundations. Argument and ethics.
And what I like most is building. Taking a messy problem, solving it and seeing the solution work in a company's day to day. When it works, the company feels it in its day to day.
Build, with AI
I build with a team of agents.
I split the work across AI agents: research, code, tests and review. The final call is always mine.
I've used AI since the first ChatGPT to go further with what I already do as a developer, not instead of it.
Learn more
- TECHNIQUES
- Multi-agent orchestrationRAGMCP and tool callingStructured outputsEvals and LLM-as-judge
Lower cost. The right model for each task, lean context and caching.
Faster. Agents in parallel when tasks don't depend on each other.
Fewer hallucinations. Answers with sources, checked and validated before they reach me.
Some I use every day; others I'm still going deeper on.
See all techniques
Best known
- RAG
- Chain-of-Thought
- Few-shot
- Prompt chaining
- Tool use / function calling
- MCP
- Structured outputs
- Hybrid search
- Reranking
- Embeddings
- Knowledge graph
- LLM-as-judge
- Guardrails
- Prompt caching
- Fine-tuning / LoRA
- Ollama
- Self-consistency
- ReAct
- Query rewriting / HyDE
- Evals / golden dataset
- Observability
- Context engineering
Instructions and structure
- Closed-world lists
- Instruction hierarchy
- Explicit abstention
- Delimited source of truth
- Entity normalisation
- Glossary in the prompt
- Scope and out-of-scope
- Clarifying questions
- Premises section
- Fact vs inference vs uncertainty
- Per-claim confidence
- XML tags
- Document first, question last
- Positive instructions with reasons
- Premise check
- Don't suggest the answer
- Current date injection
- Explicit units and formats
- Answer prefill
- Answer templates
- Nullable fields
- Enums
- Short answers
- No ALL-CAPS
- Principles over details
- No invented specifics
- Staleness flags
Reasoning
- Step-back prompting
- Plan-and-solve
- Least-to-most
- Self-ask
- Tree of Thoughts
- Reasoning budget
- Extended thinking
- Classify before answering
- Recitation
Examples
- Dynamic few-shot
- Few-shot with refusals
- Labelled negative examples
- Diverse examples
- Order and label bias
- Example leakage
RAG
- Mandatory citations
- Contextual retrieval
- Multi-query
- Parent-document retrieval
- Sentence-window retrieval
- Metadata filtering
- Recency filter and dedup
- Lost in the middle
- Relevance filter
- Answerability classifier
- Score threshold
- Self-RAG
- CRAG
- FLARE
- Structured queries for exact data
- Numbered chunks with metadata
- Source conflict reporting
- Quote-first
- Grounding
- RAG metrics
Verification
- Chain-of-Verification
- Validate and repair
- SelfCheckGPT
- Semantic entropy
- Self-critique
- Judge rubrics
- Position bias
- Cross-model voting
- Model debate
- Span-level verification
- Verification by execution
- Guardrail validators
- Abstention calibration
- Logprobs
Agents and tools
- Evidence before success
- Tools over memory
- Strict function calling
- Good tool descriptions
- Tool argument validation
- External state as truth
- Verification checkpoints
- Clean context for review
- Context poisoning
- Instruction drift
- Sycophancy
- Stop sequences
Local models
- num_ctx
- Constrained decoding
- Abstention fine-tuning
- Correct chat template
- Low temperature
- min_p / top_k
- Fixed seed
- Sandwich prompting
- Right quantisation
- Instruct models
- RAFT
- Distillation
- Transcribe before interpreting
Process and measurement
- Regression evals
- Meta-prompting
- Prompt optimisation (DSPy, TextGrad)
- User feedback loops
Result
Live, and working.
Then I ship it and adjust until people use it.
Learn more
- TYPES
- Back officesWeb and mobile appsIntegrations and APIsData dashboardsAI toolsMarketing and acquisition
I can help if you…
Quality
Built to last.
I leave it documented and tested, so anyone can pick it up.
Learn more
What kind of projects do you take on? Consulting, and projects that need to be done well: web and mobile products, back offices, integrations and AI tools.
What if you disappear? The project can't depend on me. Everything is documented, tested and easy for someone else to continue.
Doesn't AI write bad code? It can, if nobody reviews it. That's why nothing goes to production without review, tests and security.
How much does it cost, and how long does it take? It depends on what you need. Tell me about the project and I'll reply with a path and next steps.
What about confidentiality? It stays between us. I sign an NDA before seeing anything sensitive and follow NDAs and contracts to the letter: I do not work for competitors of the companies I work with. Of the work I did for other companies, I only show what is public or cleared, and nothing that affects their security or a contract.
Contact
Tell me about your project.
If this sounds like what your project needs, let's talk.