Luis Lirola
Forward Deployed Engineer / Applied AI Engineer
- Wrocław, Poland
- English — professional working proficiency
- lirolaluis@gmail.com
- lirolaluis.com

Professional Summary
Customer-facing engineer and solutions architect with 14+ years in software engineering, from discovery and architecture to full-stack implementation and production delivery. Combine enterprise experience in telehealth, finance, education and cloud modernization with ownership of live products. Build LLM-powered workflows, RAG and hybrid retrieval, authenticated MCP tools and human-approved agents, with explicit evaluation and integration boundaries. Use coding agents across implementation, automated checks and review.
Technical Skills
- Generative AI & Agents
- LLM APIs (OpenAI, Amazon Bedrock), prompt engineering, context management, structured outputs and extraction, tool/function calling, bounded agent orchestration, guardrails and human-in-the-loop approval
- RAG & Evaluation
- Retrieval-augmented generation (RAG), GraphRAG, knowledge graphs, ontologies, metadata modeling, document ingestion, parent/child chunking, multilingual embeddings, vector search, hybrid retrieval (RRF), pgvector, citations, retrieval benchmarks, regression and adversarial evaluations
- Cloud & Platform
- AWS (EKS, Lambda, S3, Aurora, DynamoDB, Bedrock), Microsoft Azure, Entra ID, Docker, Kubernetes, CI/CD, cloud migration, legacy coexistence and production delivery
- Product Engineering
- Java, Spring Boot, Python/FastAPI, Angular, React, TypeScript/JavaScript, SQL & NoSQL databases, data modeling, schema migrations and background jobs
- APIs & Integrations
- MCP servers and tools, REST APIs, OAuth 2.0/PKCE, enterprise identity, FHIR R4, event-driven integration, outbox/dead-letter patterns, idempotency and authorization boundaries
- AI-assisted Delivery
- Codex, Cursor, task decomposition, repository instructions, agent-driven implementation, automated checks, code review, regression analysis and release verification
Core Expertise
- Customer discovery and requirements
- End-to-end product and integration delivery
- Applied AI: LLM workflows, RAG, agents and evaluation
- Technical leadership and production engineering
Professional Experience
DataArt Solutions, Inc.
Lead enterprise discovery, solution architecture and technical planning with clients and product owners.
- Provided technical direction across three teams on a live telehealth platform: HLD/LLD, data modeling and release coordination.
- Designed phased modernization from Azure / SQL Server / .NET toward AWS, Aurora and Java, with domain boundaries, legacy coexistence, shadow mode, transition bridges, outbox and dead-letter handling.
- Shaped CMS integrations, metadata, localization and delivery planning for a learning-content platform; its first release reached staging.
- Contributed to content knowledge graphs: metadata modeling, relationships, topology and ontologies for enterprise content workflows.
Center of Open Systems and High Technologies
Selected Projects
Pets pets pets
Own product discovery, architecture, full-stack implementation and delivery across retail, a B2B partner platform, business-data tools and AI-assisted operations. pets-pets-pets.com.
- Built retail checkout and operations, plus a separate B2B partner platform with company pricing, payment terms, atomic stock reservations and procurement.
- Integrated BaseLinker, Taxxo and Allegro for sales, inventory, finance and advertising reports, preserving source freshness and account boundaries.
- Built a Python/FastAPI knowledge service with PDF ingestion, page-aware parent/child chunking, multilingual embeddings, pgvector search and lexical/vector fusion using RRF.
- Exposed authenticated MCP search and read-only business tools through OAuth/PKCE, reusing reporting services and preserving source freshness and account boundaries.
- Built an SMM agent and editorial workspace for weekly planning, queued text/script generation and production briefs, with version history, feedback and exact-version human approval.
- Kept generation outside long database transactions and guarded content versions and approval state against stale worker results and concurrent edits.
TechnologiesJava, Spring Boot, Angular, React/TypeScript, Python/FastAPI, pgvector, hybrid RAG, MCP, OAuth/PKCE
Healthcare Referral Intake Assistant
An AI intake desk for a fictional clinic: structured referral extraction, grounded rule retrieval, administrative checks and a local FHIR R4 Task workflow, evaluated on synthetic data.
- Implemented structured LLM extraction with deterministic labeled-field verification; separated fixture and connected-model execution for repeatable testing.
- Built lexical, model-vector and hybrid retrieval with access/effective-date filtering and source citations; exposed rule search through a read-only MCP boundary.
- Compared a fixed pipeline with a bounded agent allowed to choose read tools. Kept exact human approval, draft hashes and durable commit idempotency in server code.
- Evaluated 40 synthetic referrals: lexical recall@5 0.95 versus 1.00 for model-vector/hybrid retrieval. Recorded an injected-field failure, regression fix and holdout contamination.
- Compared three live agent runs with the baseline: identical draft hashes, higher token use. Used the result to make the agent tradeoff explicit.
TechnologiesJava, Spring Boot, Angular, LLM APIs, embeddings, pgvector, FHIR R4, MCP, synthetic evaluations
Investivoq
Live investment research platform: strategy catalog, subscription access, alerts and ingestion/publishing tools. Research only. investivoq.com.
- Shipped the Java/Angular product and integrated a Python research engine, with strategy publishing, subscription access and alert workflows.
- Integrated external market-data sources and ingestion/publishing tools for a live investment-research product.
TechnologiesSpring Boot, Angular, PostgreSQL, Python
Further selected work is available at lirolaluis.com/projects.