Hello, I'm Bernardo Olisan

I engineered and optimized a groundbreaking API interfacing with a large language model (LLM), which included both Proactive and Reactive AI systems. This involved developing a Reactive AI system using RAG, significantly reducing coaches' response time by 70% through expedited client data access. Additionally, I personally centralized databases, resulting in an impressive 90% improvement in system speed for real-time data retrieval. Furthermore, I innovated the Proactive AI to process Intercom webhooks, enabling prompt alerts for immediate action, seamlessly integrated with Slack messages.

MongoDB

Python

JavaScript

I co-founded StandardsAI with CRESE, aiming to simplify compliance with NOM 035 STPS 2018, a mandatory standard for Mexican companies. We developed a system to automate interviews and assist companies in meeting their compliance requirements independently. Additionally, I created a survey application and reporting feature to streamline processes for both CRESE and our clients.

MongoDB

Nextjs

JavaScript

Nodejs

As a personal project, I implemented the NEAT (Neuro Evolution of Augmented Topologies) algorithm from scratch using C++ and advanced mathematical principles, following the official research paper. I modified the algorithm to incorporate a unique better-worse system, enhancing learning efficiency and accelerating convergence by 10 times. Additionally, I open-sourced the project on GitHub to foster collaboration and knowledge sharing.

C++

Python

I implemented the Transformers architecture from scratch before LLMs gained widespread popularity because I believed in the potential of the first GPTs. I even shared videos about it on TikTok. Using this architecture, I developed an Automatic Speech Recognition (ASR) system, which was successful. However, I encountered issues due to limited computational power. To overcome these challenges, I required a dataset 50 times larger and needed to train it for more days.

Python

Tensorflow

As a personal project, I independently implemented the Proximal Policy Optimization (PPO) algorithm, a reinforcement learning policy-based approach. Following the official paper, I incorporated advanced mathematical principles to create a robust and efficient implementation. Moreover, I shared the project on GitHub to contribute to the open-source community and facilitate knowledge sharing.

Python

PyTorch