Hugo Llach
Head of Development
Lately, I've been using AI tools to build software for all sorts of solutions. What I've found is that bringing AI into the team requires a unique set of skills. The more experience you have building different solutions with several architectures and technologies, the easier it is to know what to ask the AI, to challenge its suggestions, improve them, review its work, and ask for corrections. To truly integrate AI, you have to adjust your methodology and learn to be very precise in defining tasks. This is where new practices like Prompt Engineering come in, along with good old-fashioned practices like breaking down architecture and goals into smaller units. Perhaps the toughest challenge is keeping the AI focused on long or complex projects, where new Context Engineering practices are showing promising results.
The latest Stack Overflow Developer Survey shows more developers are willing to use AI tools, but it also reveals a sharp drop in trust for the code they generate, with the most experienced developers becoming the most skeptical. We've also seen executives become more hesitant to invest in AI for their dev teams. Yet, big companies with brilliant people are launching new AI tools at a mind-blowing pace. Why the contradiction? Because like any new tech that promised to be a silver bullet, AI agents don't work magic on their own. You can't just drop them into a team that isn't ready. The secret is to seize the opportunity now by introducing these tools gradually, improving your methodology, and training your team. The magic isn't in a specific tool, but in how you learn to use them. That's where you should focus.
Prototypes Built with AI
Here are some prototypes I've built with AI as my sole team member—helping me define the concept, user experience, architecture, build, testing, debugging, searches in forums and documentation, infrastructure setup, and deployment on a Kubernetes cluster in the Google Cloud.
FutureLearn Academy
Multilingual chatbot assistant. Ask any questions about the school, courses, instructors, and careers in any language. Note how the bot guides the conversation. Tip: Try to steer the bot into off-topic conversations. Or try offensive criticisms about the school, courses, instructors, prices, etc.
More details
Polyglot RAG (Retrieval-Augmented Generation) chatbot assistant. Built using a Gemini-Flash model backend with a RAG pipeline. The vectorized knowledge base about the school is stored in MongoDB Atlas Cloud Service.
Vector Reel Movie Recommender
Movie recommender based on a semantic vectorized version of the MovieLens Database. Describe what type of movies you like, and get recommendations for your next weekend. This technique is very useful for offering products or content when historical user information is unavailable.
More details
Uses sentence-transformers to create embeddings from movie plot summaries. Recommendations are calculated using cosine similarity between the user's query vector and the movie vectors. It can be easily complemented with machine learning models to consider other non-semantic aspects in the product selection.
PureSocial
Social media deeply integrated into Genesys Cloud for text chatting, video calls, chatbots, and more. This is a real product whose development I led technically from its earliest stages. Puresocial is now part of Genesys' offering for its premium clients worldwide.
More details
Not coded with AI. Almost all of my projects have non-disclosure clauses or are complex backend implementations with no end-user interface. So I decided to include Puresocial here, taking advantage of its public information. You can sign up for a trial on Appfoundry.
ACME University Certificates
Anti-tampering certificates issued by the University and secured by the Polygon blockchain. You can generate your certificate, download it, and then validate its integrity.
More details
Integrates with MetaMask for wallet interactions. The certificate's hash is stored on the Polygon testnet (Mumbai) via a custom smart contract.
Launchpad with MEV-Resistant Token
Deploying a token on Solana devnet that can be bridged to Polygon testnet. It includes MEV resistance by using private relays or bundles during deployment to avoid sniping.
More details
Utilizes Solana's Web3.js for token creation and deployment scripts. MEV-resistance is explored through integration with Flashbots or similar private transaction services.
Relevant Experience
I've spent 20+ years leading teams building high-traffic, asynchronous software for mobile operators and contact centers, and automating system administration tasks for retail and finance companies.
2025-
Head of Development
Eternal Technology
2024-2025
Independent Development Consultant
AI, Blockchain and others.
2015-2023
Development Director
Sixbell - Contact Center solutions
2012-2015
Independent Technical Consultant
Falabella, Habitat, Netline, others
1999-2011
Head of Development, Projects, R&D
Sixbell - Telecommunication Solutions
1998-1999
IT consultant
Codelco - System automation
1995-1998
Unix support and Administration
Sonda Chile - Digital Equipment Corporation
Let's build the future, together.
I'm available for consulting and workshops. Let's talk.
Send an Email