Building AI systems that turn intent into working, reviewable software.
Syntran = Synthetic Translation: AI-assisted engineering that converts ideas, specs, and research into structured, auditable systems.
Syntran Labs is the public AI engineering portfolio of Leonardo Sigales — an engineer with 20+ years in IT, transitioning from Data & Analytics leadership into AI Engineering, building on hands-on GenAI experience at Sabre.
- Designed and shipped LLM chatbot solutions in production using Dialogflow CX, prompt design, conversation workflows, and response-quality iteration
- Built GenAI workflows to improve user interaction at scale
- Owned end-to-end systems on Google Cloud, including BigQuery, Vertex AI, and a production ML model for fraud detection
- Spent 20 years bridging technical and business teams — the skill AI products live or die by
Why this matters: most AI engineering demos break when they meet production constraints. Two decades of operating real systems — data quality, incident management, stakeholder pressure, documentation, and security boundaries — are the foundation these projects are built on.
Published work now spans Learning Lab, Paper Lab, and Systems Lab: educational engineering workflows, research-to-code implementations, and production-grade AI engineering systems.
| Project | Status | Track | What It Demonstrates | Stack |
|---|---|---|---|---|
| syntran-aieos | Published | Systems Lab | AI Engineering Operating System for Claude Code: governed agents, repeatable skills, explicit permission gates, and a self-measuring telemetry layer. Windows-first. | Claude Code · Markdown · PowerShell · Python |
| learn-spec-driven-dev | Published | Learning Lab | Spec-Driven Development, executable specs, pytest, Red-Green-Refactor, dependency injection, and responsible AI-assisted engineering | Python · pytest · OpenSpec |
| paper-rag-graph-4-datasets | Published | Paper Lab | Research-to-code implementation of a Graph RAG pipeline for explainable dataset discovery | Python · Jupyter · NumPy · pandas · matplotlib · pytest · GitHub Actions |
| paper-eca-llm-hypothesis-workflow | Early Incubation | Paper Lab | Uses SYNTRAN AIEOS to test whether LLM-assisted scientific workflows can produce falsifiable, reproducible, non-overclaiming hypotheses. ECA as a governed, reproducible testbed. | Python · SYNTRAN AIEOS · LLM APIs |
| Permission-aware RAG service | In Progress | Systems Lab | Secure retrieval: access control at retrieval time, citations, traceability, evals | Python · FastAPI · vector DB · LLM APIs |
| LLM agent with tool use | Planned | Systems Lab | Agent orchestration: planning, function calling, guardrails, failure handling | Python · LLM APIs · structured outputs |
| Eval & observability harness | Planned | Systems Lab | LLMOps: automated evals, prompt regression testing, tracing, cost/latency monitoring | Python · pytest · tracing tools |
| Document-processing pipeline | Planned | Systems Lab | Applied GenAI: extraction → validation → structured output from messy real-world documents | Python · OCR · LLM APIs |
| Track | Purpose |
|---|---|
| systems-Lab | Production-oriented AI engineering systems: AI engineering operating systems, governed agents, repeatable skills, permission governance, observability, secure RAG, and operational readiness |
| paper-lab | AI/ML research papers turned into simplified implementations, experiments, and engineering notes |
| learning-lab | Public catalog of self-contained educational engineering repositories — the front door for learning projects like learn-spec-driven-dev |
- Ideas should become structured artifacts
- AI-assisted work must remain human-reviewable
- Documentation is part of the engineering output — every project ships with architecture, trade-offs, and limitations
- Security and privacy are designed in from the start, especially in retrieval and agent systems
- No demo-ware: projects include evals, failure modes, and operational concerns — or they say honestly that they do not
AIEOS: governed agents, skills, telemetry [v1] Secure RAG & permission-aware retrieval
LLM agents: tool use, orchestration, guardrails Evals & LLMOps: testing, tracing
AI-assisted & spec-driven software delivery Research-to-code exploration
Created and maintained by Leonardo Sigales
Open to conversations about applied AI, GenAI in production, and engineering leadership.