Sergio Gómez Martín

A hands-on lab exploring the LLM lifecycle, from a small GPT with PyTorch to adapting open-weight models.

• Foundations: data preparation, attention, and GPT architecture with PyTorch.
• In progress: exploring Qwen3 and an SFT experiment for tool-calling with MLX.
• Next: compare the base and fine-tuned models using the same evaluation.

• PyTorch
• Qwen3
• MLX
• SFT
• Tool-calling

A financial-health platform built for the Embat challenge at HackSpain 2026, designed to help CFOs understand where a company is heading from its cash-flow trajectory.

• Produces a monthly 0-to-100 health score that combines the current level with its trajectory.
• Distinguishes isolated cash-flow dips from sustained deterioration and explains the drivers behind each change.
• Turns the score into alerts, working-capital offers, ranked actions, and plain-language insights from AI agents.

• Python
• FastAPI
• React
• DuckDB
• Financial health

Academic project “Autonomous Architecture for Generation and Coordination of LLM Agents”: an autonomous multi-agent architecture that breaks down complex requests, generates specialised LLM agents, and orchestrates dependencies and tools.

• Decomposes complex requests and dynamically generates specialised LLM agents.
• Coordinates task dependencies and tool use.
• Built with FastAPI and Docker, with automated tests.

• LLM agents
• Agent orchestration
• FastAPI
• Docker
• Automated tests

A serverless data platform on AWS using CDK, Glue, Lambda, Apache Hudi, and DynamoDB.

• AWS CDK
• AWS Glue · Lambda
• Apache Hudi · DynamoDB

• Apache Iceberg
• AWS
• CDK
• Glue
• Lambda
• Hudi
• DynamoDB