Juan Pablo Sotelo

Developed solutions utilizing generative AI text models, working both with ChatGPT and other proprietary models as well as OSS models to match compliance requirements, diving into the end-to-end app development process with LLMs.

Description: designed a tech support Chatbot for an application with more than 1M+ monthly users. The chatbot had access to up-to-date documentation of the application best use practices, common error handling and procedures to simplify the user experience. Chatbot's latency was optimized to less than <1s per generation.

Re-architected the search stack for an e-commerce platform with +100k daily search users by introducing hybrid search for retrieval, combining lexical BM25 with semantic multimodal embeddings, and created a re-ranking model that leveraged user data interactions to re-rank the results. These two upgrades resulted in a +13% session conversion uplift and a 11% time-to-product reduction. The re-ranking part of the project included CI/CD for model automatic training and deployment.