Carlos – Python, AWS, OpenAI

Enterprise-grade autonomous AI agent that automates complex ESG data analysis workflows for 50,000+ companies globally. The system processes unstructured data from multiple sources, makes intelligent decisions, and generates comprehensive sustainability reports, reducing analysis time from days to minutes.

• Architected end-to-end agentic workflow using LangGraph, implementing sophisticated state management and decision trees that handle 15+ different tool integrations;
• Engineered advanced RAG system with hybrid search (semantic + keyword) achieving 94% retrieval accuracy across 1M+ documents;
• Designed and optimized 50+ complex prompt chains with self-correction mechanisms, improving agent task completion rate from 60% to 92%;
• Implemented dynamic tool selection logic allowing the agent to autonomously choose from 20+ specialized tools based on context;
• Reduced operational costs by 70% through intelligent LLM routing (GPT-4 for complex tasks, smaller models for simple ones);
• Built a comprehensive evaluation framework with custom metrics to monitor agent performance and detect hallucinations.

Comprehensive AI and Data Science education program training 200+ professionals annually in cutting-edge technologies. The program combines theoretical foundations with hands-on projects, preparing students for real-world AI implementation in enterprise environments. Students achieve an average 40% salary increase post-graduation.

• Designed and delivered advanced AI curriculum covering LLMs, RAG systems, and agentic workflows to 200+ students;
• Created hands-on projects simulating real enterprise challenges, with students building production-ready AI applications;
• Developed innovative teaching methodology combining live coding, interactive workshops, and personalized mentoring;
• Achieved 4.9/5 student satisfaction rating and 95% course completion rate through engaging content delivery;
• Mentored final master projects.

An AI-powered meditation app that creates personalized audio meditations in real-time. Users describe their current emotional state through voice, and the system generates unique, context-aware guided meditations with natural-sounding AI voices, serving 500+ beta users with a 4.8/5 satisfaction rating.

• Built a conversational AI system capable of conducting empathetic voice interviews to assess users’ emotional states and preferences;
• Engineered a multi-modal pipeline integrating Whisper for voice-to-text and ElevenLabs for text-to-speech with latency under two seconds;
• Designed a prompt engineering framework to generate diverse and contextually appropriate meditation scripts;
• Implemented a user preference learning system that adapted meditation styles based on historical feedback.

An AI-powered news intelligence system that processes 100,000+ daily articles to identify ESG-relevant content and automatically group related stories. The system serves as the backbone for real-time sustainability monitoring across global companies and industries.

• Fine-tuned LLaMA-2 model using LoRA/QLoRA techniques for news categorization, achieving 89% accuracy with 10x faster inference;
• Developed a hierarchical clustering algorithm combining embedding similarity and LLM reasoning to group related articles;
• Reduced manual screening labor by 85% through a multi-stage relevance detection pipeline using sentiment analysis and entity recognition;
• Optimized model serving with vLLM, handling 1,000+ requests/second with p99 latency under 200ms;
• Implemented an active learning loop that continuously improved model performance based on human feedback
• Built a real-time monitoring dashboard tracking system performance and data quality metrics