Public-safe summaries of the work I can discuss, and the role I do it in. Each project links to the closest live lab on this site, and says how close it is.
Selected projects
01
Multi-agent system that executes multi-step work across business tools
Multi-step work that spans APIs, databases and external services is hard to hand to a single model call. I designed and built a multi-agent system that reasons, plans and carries that work out.
Result Carries multi-step tasks across APIs, databases and external services, with fewer failure cases in complex workflows.
How I built it
The agents use tools and coordinate their actions, adapting to intermediate results instead of following a fixed script. Long-running tasks keep their memory and context between steps.
This work is described in my CV. I'm preparing a public-safe case study that explains the architecture without exposing client systems or private data.
RAG system that answers from structured and unstructured sources together
Useful answers often depend on both structured records and unstructured documents. I built a retrieval-augmented generation system that pulls from both in real time.
Result Answers in real time, grounded in retrieved context from structured and unstructured data.
How I built it
Embedding and vector-search pipelines feed the model. I tuned chunking and retrieval to improve relevance and cut hallucinations, checked groundedness and answer quality against offline benchmark sets, and monitored response quality in production.
This work is described in my CV. I'm preparing a public-safe case study that explains the retrieval and evaluation decisions without sharing private source material.
Vertex AI pipelines that take models from preprocessing to managed endpoints
Training a model is only part of the job; it also has to run reliably in production. I designed and deployed Google Cloud Vertex AI pipelines that cover data preprocessing, training, inference and endpoint management.
Result Serves real-time inference inside production systems, from data preprocessing to a managed endpoint on one repeatable path.
How I built it
The ML workflow is wired to the cloud infrastructure and operating practices around it, so a model moves through preprocessing, training and deployment on one repeatable path and scales from there.
This work is described in my CV. I'm preparing a public-safe case study focused on the pipeline, infrastructure, and deployment decisions.
Real-time matching system built for noisy, incomplete data
The hard part of this system was ambiguity: inputs that are noisy, incomplete or contradictory. I designed a modular real-time inference pipeline that combines computer vision, data ingestion, NLP processing and contextual matching.
Result Matches in real time while handling ambiguous and incomplete inputs, with each component scaling independently.
How I built it
A vision model turns faces into embeddings, public records are normalized into comparable profiles, and the matching logic weighs partial evidence instead of forcing a yes or no. Vision, ingestion and inference are separate services, so each scales on its own.
This work is described in my CV. Any public case study will focus on the engineering architecture and responsible system boundaries, not personal or sensitive source data.
Remote from Quito, Ecuador · Capmation is based in the Twin Cities, MN, US
I design and deliver production AI and software systems, from the first technical direction through implementation, deployment and the improvements after. The work covers agentic AI, RAG, backend services, cloud delivery, data infrastructure, and the engineering practices that keep those systems dependable.
My AI work sits inside real business processes: LLM-powered applications and workflows wired into the APIs and databases a team already depends on.
When I design multi-step agents, retrieval pipelines and automation, I set clear tool boundaries up front so each piece has an owner once it's running.
The Python and .NET backend services behind those systems also handle real-time data processing, and I deploy them across Google Cloud, AWS and Azure.
Vertex AI, Docker, Kubernetes and CI/CD are how I carry models and services from development into production at scale.
I work directly with cross-functional teams to turn business needs into systems we can ship, while leading delivery and mentoring engineers.
Blockchain engineering practice
My professional background is software and applied AI. These independent labs show the blockchain skills I am developing for junior blockchain roles, separate from the client projects above.
Solidity analysis
Explore static analysis findings on sample contracts. This is a learning tool, not a professional security audit.