<aside> 👨💻 Helping people get back to important work @ ToolCharm
<aside> 🏫 Honors CS @ UCF
<aside> 💼 Prev. Engineering @ Morgan & Morgan, VianAI, Tompkins Robotics
<aside> 📩 Email: [email protected] Twitter: @mark_bruckert Github: @mbruckert
<aside> ❇️ ToolCharm
ToolCharm is the startup that I co-founded.
The mission of ToolCharm? Helping people stop worrying on work about work (i.e. the back-and-forth about tasks & scheduling, tracking down information, data entry, etc.) and get back to the work they care about.
Check out our website:
ToolCharm | Supercharge your work with AI that remembers
<aside> 🪨 GuideStone [Project for Stanford’s TreeHacks 2024]
Introducing GuideStone! We each have our own educational path, and GuideStone understands that. We build out a graph for each and every user that expands as you learn automatically. For example, if you want to learn integrals - but haven't yet learned derivatives, our platform will first generate a lesson on deviates to make sure you have a full grasp on the topic.
I mentioned lessons, but you might be wondering.. what does a lesson look like on GuideStone? Well, every lesson comes complete with a video & quiz to test your understanding. How is that better than the earlier mentioned education platforms? Well... Generative AI of course!! Every video on our platform is generated from scratch by GPT-4. We'll dive into how a bit later, but the ability to create videos on topics for a specific user allows us to do some really special things like modifying example animations to be about things you care about and resonate with like hobbies/interests/passions.
I also mentioned that GuideStone was self-improving. How does that work? Well, we incorporated a combination of eye-tracking & quizzes to modify your generated videos based on what kind of videos work to better your understanding. While you watch the video, GuideStone is gathering data on what parts of the video you are paying the most attention to without any required input from you. Then, when you finish the video, we use this eye-tracking data to both surface general lesson content, but also specific content we think you might have missed during the video. We use an agentic architecture to then analyze this data, and plan a form of action on how to create better videos (one's which perform better on attention & quiz score) for you in the future.
Here’s a technical overview thread by my teammate of how it all works:
Owen Burns on Twitter / X
<aside> 🧠 GPTeach [Project for FIU’s ShellHacks 2023]
1st Place - Microsoft AI Challenge
GPTeach is your complete video-based, generative AI tutor that can generate complex animations, 2D and 3D graphs, formulas, and so much more. The goal? Visually explain any topic to you, in addition to our TTS generation which explains the topic as well as explains what is going on in the visuals
Additionally, along with generating these videos based on prompts or questions, GPTeach generates a question and a set of answer choices that allows you to test your learning from the video. Because we know that LLMs can hallucinate answers, we cross-reference the chosen answer choice with Wolfram Alpha to ensure it is correct. If the chosen answer is incorrect, GPTeach will generate an explanation of why it is wrong as well as a video visually presenting what is wrong in the user's logic and why the correct answer is the better option.
We built a GPT-4 agent that is in charge of orchestrating a storyboard of the video based on the user's prompt/question. We use a Pydantic-based output parser which forces the LLM to constrain it's output to a strict JSON format containing an array of visual and auditory prompts. We then use parallelized setup to execute Google Cloud TTS and a custom agent which generates Manim code. This custom GPT-4 agent utilizes a vector (Chroma) database which we created by scraping the documentation for Manim to run a similarity search to pull relavent documentation on generating the section of the video. The agent is then tasked with generating valid Manim code which we accomplish through a combination of various custom output parsers, which it uses to compile the animations and create the video snippets. From here, our code combines the visuals with the audio, time-matching them to ensure succinct and accurate explanations. At the end, we use ffmpeg to combine all of the video sections into one output file which we present in the web UI.
<aside> 💡 Knight Hacks VI [Organized hackathon for 800+ people]
I am a lead organizer for my university’s hackathon, Knight Hacks. Knight Hacks IV occurred October 6-8th at the University of Central Florida and our annually held hackathon in the top 10% of hackathons by size. The event required us to raise over $45,000 from sponsors like Microsoft, Morgan & Morgan, Siemens Energy, Lockheed Martin, and more. I also build the website for advertisement and registration for the hackathon!
Success Hacks: 700 Students Attend UCF Knight Hacks’ Largest Hackathon | University of Central Florida News
Some really cool projects were created at the 36-hour hackathon from tactile haptic wearable systems to AI Powered Interview Prep in VR.
<aside> 📄 IdeaSleuth [Project for Pinecone Hackathon 2023]
For the PineCone hackathon, my teammate and I created IdeaSleuth! Our project takes an idea description, finds and reads patents from across the globe that are relevant (in any language!), and generates a PDF with the related patents, a detailed analysis of the IP landscape, suggestions to improve your idea, and even a score or how patentable your idea is!
IdeaSleuth takes a description of an idea from the user and uses an LLM-powered agent (built with Langchain) to take this description and convert it into a series of SQL queries which is used to search the BigQuery database of international patents. Once relevant patents have been found, IdeaSleuth scrapes the Google Patent page for the patent in order to get the PDF, and loads all of the PDFs into our Pinecone vector database. From here, we use a GPT-4 agent to run a similarity search on our Pinecone database and answer the pre-set selection of questions, as well as assign the idea on a rating of how patentable it is. Once all of this information has been written, we use the reportlab python library to generate a stylish PDF which makes it easy to quickly consume all of the analysis about your idea and relevant IP. The front end for the application is built with React and hosted on Vercel.
<aside> 💪 Muscle Memory [Project for USF’S Hackabull Hackathon]
1st Place - Overall
Created an AI-based Personal Trainer which could interact with SQL databases and navigate the internet in order to store information about users as well as prevent LLM hallucinations by empowering it with real-world data. Utilized Ionic and React for frontend and Python, Langchain, and PostgreSQL for backend.
<aside> ⚖️ Legal Hackathon [Lead Organizer & Workshop Leader]
Co-organized a hackathon with only three weeks of planning time, the first event that Knight Hacks planned for a sponsor.
I also led a workshop for dozens of participants at the event that taught participants how to build an agent powered by OpenAI GPT-4 and LangChain that utilizes the Spotify API to become your ultimate music assistant. It knows your Spotify history, retrieves information about songs/artists/albums, can create custom playlists, and more. The audience first learned about LLMS, tools, and agents and then followed along as I built the Spotify agent - all in under an hour!