Rodrigo – LLM, RAG, LangChain

An advanced data intelligence platform used daily by credit analysts at a leading US corporate credit institution. The platform integrates multiple data sources and databases with automated ingestion, preprocessing, metadata extraction, and agentic AI capabilities to provide analysts with centralized, actionable data.I focus on the multi-agent conversational AI layer built on top of this system.

• Designed and built a Redis-backed reference-ID cache for the agent's tool-call pipeline, reducing inference time and payload size.
• Built an isolated jq-based calculation tool for sandboxed compute over cached datasets.
• Diagnosed and fixed a Bedrock prompt-cache invalidation bug, increasing cache-served tokens and reducing rewritten input tokens.
• Extended the Prompt Gallery backend from admin-only access to user-authored personal prompts with per-user isolation.

An AI-powered automation platform supporting back-office operations and helping streamline internal business processes for a large Canadian telecommunications provider serving millions of customers across the country. I was part of a team that developed and maintained a machine learning system that automatically interprets service tickets and extracts meaningful insights, significantly reducing manual processing time and improving operational efficiency.

• Aided in the architecture and deployment of an end-to-end LLM-based ticket classification system on GCP using Python and FastAPI.
• Implemented a RAG pipeline combining vector similarity search with LLM reasoning for ticket-to-defect matching.
• Designed a high-performance vector search system using Turbopuffer to retrieve relevant historical tickets.
• Designed distribution-based evaluation metrics to validate defect prioritization quality.
• Developed an MCP server enabling natural language querying of tickets through text-to-SQL translation.
• Built an asynchronous Firestore logging system with request-context injection for API tracing in GKE environments.

A real-time AI-powered system that processes customer messages to automate sales operations, helping B2B sales representatives manage customer portfolios and process orders efficiently. Under my leadership in AI and backend engineering, the company grew 30x in revenue in 2024.

• Led the development of the AI-powered sales agent using Golang and Python for real-time B2B sales automation.
• Built a temporal workflow system that managed conversation state and maintained context across interactions.
• Created a perception engine that processed images, audio, spreadsheets, PDFs, and text messages in parallel.
• Implemented a multi-tenant platform allowing each organization to customize agent behavior.
• Integrated multiple LLM providers into the agent pipeline, including OpenAI, Azure, Anthropic, and Amazon Bedrock.
• Developed a high-performance LLM-based pipeline that created or updated a shopping cart in 15-30 seconds on average.
• Built an event-driven cart management system that parsed and executed commands from natural language.
• Developed a cart item matching system combining full-text and semantic search across textual data and vector embeddings.

An AI application platform focused on developing and deploying machine learning solutions across their full lifecycle, from initial development and experimentation to production deployment and ongoing improvement.

• Led end-to-end development of conversational AI for sales and customer service using Python and the Rasa framework.
• Implemented the complete AI workflow including preprocessing training data, training and testing the model, and conducting unit and integration tests.
• Developed intent classification and entity extraction features for quotation identification and client data extraction.
• Maintained the bot’s CI/CD pipeline and evaluated system compatibility with LLM models.
• Designed and implemented a computer vision system in C++ using OpenCV and Torch for monitoring auto parts packing processes.

A smart data platform for autonomous vehicles, focused on processing and managing vehicle data to support the development and operation of autonomous driving systems. Led the integration between a Computer Vision system for Centerline Estimation and EPOS with real-time constraints.

• Worked on the SDAV project and led the integration between a computer vision system for centerline estimation and EPOS with real-time constraints.
• Developed a port of EPOS for RISC-V architecture, focusing on hardware mediators and networking protocols.
• Designed and implemented SmartData and TSTP networking solutions for autonomous vehicle communications.
• Conducted literature review and implemented a convolutional neural network for lane detection and segmentation.
• Led integration between the Python-based lane detection system, C++ SmartData management, and C++ EPOS real-time car software.
• Engineered a high-performance UDP packet system to connect AI capabilities with real-time vehicle subsystems.