I am motivated to apply to one of the most groundbreaking industries of our time, to contribute to the efforts of utilizing AI for the betterment of society, and contributing against any potential alignment threat, by any capacity I'm entrusted, to the utmost of my abilities and without compromise. I am driven by the transformative potential of AI and automation, and how they can drive meaningful change.
Since February 2023, I have been independently working on https://librechat.ai, an enhanced ChatGPT clone with multi-user login system, multiple AI providers to choose from, and AI agency through plugins, which now has over 1000 stars and 280 forks on github.
In May, I shared my progress on reddit of reverse-engineering the ChatGPT plugins functionality before the OpenAI functions were released, and before many chat interfaces employed similar techniques. I was able to use the OpenAI API for agency of the AI with tools through LibreChat, inter-weaving into AI conversations the use of text-to-image generation with DALL-E and stable diffusion, as well as use of Wolfram, search engines, and web-scraping, as selected by the user. The post received over 100 upvotes and 20,000 views: https://www.reddit.com/r/GPT3/comments/13ggcv5/reverseengineeringchatgpt_plugins/.
Since then, I've expanded the same reverse engineering through OpenAI functions. I have working proof-of-concepts in both python and JavaScript that mimic ChatGPT's handling of OpenAPI specs, and as well as for utilizing vector retrieval for expanded/improved context through documents. As I've already shared the Clone project, I would like to share my quick MVP of a python FastAPI that serves some of these functionalities: https://github.com/danny-avila/ai-services.
As a scheduling manager who has learned to program, I have been able to create my own tools. I've developed automatic assignment and balancing tools, which I shared a video of on LinkedIn, showing over 400 real shifts for a single pay period being automated and balanced within a few minutes. This task would take me hours if not days with my previous tools and paid software solution. Due to the pricing and limitations of scheduling in the current SaaS market, I was inspired to take on this task, as my previous pre-planning and analysis methods had reached their limits. My goal is to provide a user-friendly interface for the next manager and make the backend even more flexible and robust. I'm currently studying machine learning and linear programming to advance optimization. You can see the video I shared here: https://www.linkedin.com/posts/danny-avilaprogramming-automation-scheduling-activity-7036836068393906176-TKCC?utmsource=share&utmmedium=memberdesktop.
I documented my progress, from beginning to end, on designing and scaling a legacy API into micro-services, serving up to 1000 requests per second, with less than 20 ms response time per request, and 0% error rates at this throughput under realistic test scenarios: https://gist.github.com/danny-avila/1387fef054da77737e1ce4d04172afe4. To achieve this, I utilized Postgres, express, NGINX, Redis, AWS EC2 instances, along with Pandas and NumPy for the ETL process of the legacy data. I made sure to index my data, and craft my schema and queries carefully for performance, flexibility and maintainability. I also ran stress tests with loader.io, as well as with the k6 suite, to measure performance along the way.
I'm hoping to be considered by the merit of my independent open-source work and learning, and how I've helped people use AI tools effectively, in many capacities. At my current employment, I was able to automate much of my workflow, which involves managing a schedule of over 300 people, has helped our operations immensely. This kind of work, that multiplies exponentially in value as people engage with or even indirectly benefit from created tools, is incredibly fulfilling for me, even when I don't receive any kind of compensation or recognition for it.
