Hi! I'm Leonan Vasconcelos, a senior technology, data and applied AI architect with 20+ years of experience building software platforms, machine learning systems and production-grade AI solutions across Europe and the Americas.
My background combines software engineering, software architecture, machine learning, deep learning, NLP, LLM systems, data platforms, distributed systems, geospatial workflows and operational product delivery. I focus on turning technically complex ideas into reliable systems that can be designed, implemented, validated and operated in real environments.
You can learn more about my experience and background on my LinkedIn profile.
Below are examples of work that I can share publicly and that reflect different parts of my profile: applied machine learning, deep learning, agentic AI, software architecture, data-intensive processing, reliability engineering and performance optimization.
| Project | Preview |
|---|---|
| Production-Grade Agentic RAG Architecture with LangGraph and Local LLMs Designed and implemented a trilingual (Spanish, Portuguese, English) conversational AI system for drone photogrammetry planning, combining LangGraph, local RAG, multilingual retrieval, local LLM inference and deterministic business logic. Built the pipeline end to end - intent understanding, structured extraction, retrieval, plan selection, response generation, verification and evaluation - and benchmarked five interchangeable NLP engines on the same 140 golden test cases: a TF-IDF Naive Bayes classifier trained on its own hand-built vocabulary, against four local LLMs (Llama 3.1 8B, Llama 3.2 3B and Ministral 3 3B) served by llama.cpp on a single consumer GPU. Two long-form articles document the study end to end, from the migration off keyword matching to ten measured stages of NLP optimization. This project highlights AI engineering, architecture, evaluation discipline and production reliability, not just prompt orchestration. |
 |
| AI Engineering Framework for Reliable Agentic and LLM Workflows Built a reusable framework for reliable LLM, document and verification workflows, with a runner-neutral CLI, deterministic contracts, guarded agents, automated audits, reproducible artifacts and CI/CD. This project demonstrates software architecture, testing, quality engineering and the ability to turn AI workflows into maintainable production software. |
 |
| Custom CNN Architecture, Training and Optimization for Image Reconstruction Built a custom convolutional neural network for grayscale-to-color image reconstruction on CCTV imagery. The work reflects hands-on deep learning fundamentals: model design, image preprocessing, training workflow, prediction analysis and iterative improvement of a CNN-based task, reinforcing practical machine learning capabilities beyond the use of pre-trained models. |
 |
| Algorithmic Object Detection Pipeline without Deep Learning Implemented an object detection pipeline using classical computational methods rather than neural networks. This project demonstrates algorithm design, image analysis, feature extraction, filtering and detection logic, showing strong foundations in applied computation and signal processing in addition to deep learning. |
 |
| Feature Engineering Pipeline for Visual Data Representation Developed a feature extraction workflow to transform raw images into structured numerical descriptors. The project demonstrates feature engineering, data representation and analytical preparation for downstream ranking, similarity and detection tasks. |
 |
| Feature-Based Similarity Ranking and Retrieval System Built a similarity ranking approach based on extracted feature representations, combining numerical descriptors, comparison logic and ranking into a practical retrieval workflow. This project reinforces applied machine learning concepts around representation, distance and information retrieval. |
 |
| Parallel Data Processing and Performance Optimization for Large Images Implemented a parallelized data-processing workflow for large image datasets, focusing on workload decomposition, computational efficiency and throughput. This project surfaces broader engineering strengths in parallel computing, scalability and performance optimization for data-intensive workloads. |
 |
- Software engineering and software architecture
- Machine learning and deep learning fundamentals
- CNN design, training and model evaluation
- LLMs, RAG, LangGraph and agentic workflows
- NLP evaluation and benchmarking across model families
- Feature engineering and information retrieval
- Data-intensive and parallel processing
- Testing, verification, CI/CD and production reliability