I'm Stadyn Román, an AI developer based in Guayaquil, Ecuador.
I am a senior AI developer at Banco Bolivariano, where I design agentic systems for the software development lifecycle. My main work is a spec-driven development (SDD) system for GitHub Copilot, with custom agents and skills that support teams from discovery through development and QA, connected to Jira, Figma and internal systems through MCP. The 3 teams already using it have doubled their productivity, and I also lead the operating and governance model for this way of building software. In parallel, I contribute to modernization initiatives with AWS Transform.
Before that, I spent nearly three years at Digevo taking generative AI to production: LLM-powered callbots and chatbots handling thousands of conversations a month, RAG pipelines on top of enterprise data, and automated call auditing. I also come from a research background, with five papers published by Springer and IEEE, and in October 2026 I defended my master’s thesis in Data Science.
I care about the less glamorous side of AI: making sure a system retrieves the right context, that its results can be measured, and that a team can maintain it over time.
Outside of work, I make video games in Godot, mostly for game jams, and I run my own home server.
Experience
- Jul 2026 – present
Senior AI Developer
Banco Bolivariano
- Design an agentic spec-driven development (SDD) system for GitHub Copilot, with custom agents and skills for discovery, development and QA.
- The 3 teams already using it doubled their productivity (2x), measured through throughput, cycle time and velocity; the plan is to roll it out to every development team.
- Orchestrate MCP servers for Jira, Figma and internal systems and connect them to the agents.
- Lead the operating and governance model for SDD-based development.
- Also contribute to application modernization initiatives with AWS Transform.
- Sep 2023 – Jun 2026
Generative AI Engineer
Digevo
- Designed and deployed LLM-powered sales and support callbots that handle thousands of conversations per month and cut manual work by 80%.
- Built a multilingual chatbot integrated with an e-commerce platform, improving the user experience by 80%.
- Built RAG-based conversational pipelines on top of enterprise data.
- Developed AI auditing pipelines (Terraform and Airflow) that process more than 500 calls per day.
- Optimized prompting, inference and model orchestration to keep the product robust.
- Dec 2022 – Aug 2023
Researcher / Developer
Harkay
- Developed a real-time pedestrian detection and counting system using computer vision, reaching 92% accuracy.
- Deployed it across 30+ cameras in 5 shopping malls to improve their business analytics.
- Published research on activity recognition with deep learning architectures.
Education
- Aug 2025 – Oct 2026
Master’s in Data Science
Universidad Yachay Tech
- Thesis defended in October 2026: a canton-level crime risk index and forecast for Ecuador (degree being issued).
- Published an open dataset of video game reviews and scores, plus a model that predicts scores with 72% accuracy.
- Built an open, local RAG chatbot for querying scientific papers with citations.
- Jul 2017 – Nov 2023
B.Eng. in Information Technology
Universidad Yachay Tech
- Thesis: “Illicit tweet detection using Transformers”.
- President of the Yachay Tech Student Association.
- Founding member of the Computer Science Club.
Publications
Peer-reviewed papers published in Scopus-indexed conferences and in the journal IEEE Access.
Environmental Impact of Food Products: A Data Analysis Approach Using HJ-Biplot and Clustering
ARTIIS 2023 · Communications in Computer and Information Science, vol. 1935
A Review of Techniques for Detecting Illicit Messages on Twitter
2023 IEEE Seventh Ecuador Technical Chapters Meeting (ETCM)
Accuracy Comparison of CNN, LSTM, and Transformer for Activity Recognition Using IMU and Visual Markers
IEEE Access, vol. 11, pp. 106650–106669
U-Net vs. TransUNet: Performance Comparison in Medical Image Segmentation
ICAT 2022 · Communications in Computer and Information Science, vol. 1755
Q-Learning in a Multidimensional Maze Environment
TICEC 2022 · Communications in Computer and Information Science, vol. 1648
Certifications
- 2026MLOps FundamentalsDataCamp
- 2025Microsoft Certified: Azure Data Engineer AssociateMicrosoft
- 2025NVIDIA AI Technical CurriculumNVIDIA
- 2025Developing LLM Applications with LangChainDataCamp