Cristopher – Python, LLM, AWS

It's a YC-backed US startup that uses AI role-play, coaching, and real-world feedback to build confident, capable leaders.

• Designed a reproducible LLM evaluation pipeline benchmarking outputs against a curated golden dataset.
• Built an experimentation system to compare prompt versions, model variants, and tools using human feedback.
• Architected the AI Coach agent with structured tool-calling, RAG, task scheduling, and recommendations.
• Added observability for model decisions and tool execution to enhance debugging and monitoring.
• Implemented a provider abstraction and fallback strategy to handle LLM API outages or degradation.
• Developed performance monitoring metrics, including latency, TTFT, error rates, and outage detection.

It's a platform that offers scalable language training programs for companies that want to accelerate business performance worldwide.

• Led a team of four engineers to build the Content Transformation Center (CTC) for ingesting audio, video, and image data.
• Revitalized the team by tightening the scope, clarifying the requirements, and centralizing discussions for a more focused workflow.
• Drove implementation and refinement of a content processing pipeline covering tokenization, text extraction, word sense disambiguation, classification, and text generation.
• Oversaw dictionary enhancement efforts, including manual curation, similarity-based merging, and LLM-driven definition rewrites.

It's an online learning community with thousands of classes for creative people. This project focused on the pre-existing chatbot that would chat with users to provide support and redirect them to existing flows, such as elevating the discussion to a human or issuing an automatic refund.

• Led migration of a 10,000+ LOC LLM-powered customer service chatbot from a discontinued VC-backed product, despite missing tests and deployment docs.
• Completed the migration in 3 months, restoring production service and pushing the project toward full cut-over.
• Implemented unit and integration tests to reach 85% coverage, greatly improving stability and reliability.
• Refactored and modernized the codebase with contemporary Python practices (type hints, Pydantic, FastAPI) and clear deployment procedures.
• Built a RAG system with Pinecone to ingest Skillshare HTML documentation and power question-answering.

It's a platform that offers scalable language training programs for companies that want to accelerate business performance worldwide.

• Built and streamlined a new ML pipeline with AWS Sagemaker, improving data processing and deployment efficiency in line with MLOps best practices.
• Led migration of data pipelines to Airflow from AWS Data Pipeline, boosting scalability and maintainability across SQL, Redshift, PostgreSQL, and Redis.
• Developed an advanced recommendation system and generative text models for educational content, increasing personalization and user engagement.
• Established a dynamic ML backend/API layer for model microservices and created an ML playground for experimentation.
• Drove projects in semantic search, text classification, and audio classification within a small, agile team.