Earlier this yr, Databricks launched Dolly 2.0: the world’s first actually open instruction-tuned Massive Language Mannequin (LLM). To construct off this pleasure round LLMs, Databricks hosted two hackathons, one digital, and one in-person, to kick off the annual Information+AI Summit. Information groups from across the globe got here collectively to construct distinctive use instances, functions, and strategies showcasing LLMs. Between the 2 occasions, we had over 1,000 members with over 70 tasks submitted!
Let’s take a second to showcase the successful tasks and spotlight the implausible groups who created them.
Digital Hackathon Winners
Our digital hackathon was held previous to Information + AI Summit in collaboration with the Devpost crew (particular shout-out to Michelle Mind for serving to make this such an enormous success). Contributors had 30 days to submit their tasks and will work in a crew of as much as 4 folks. Try the total venture gallery right here.
First Place: DataDM, your open supply non-public information assistant.
Gus Eggert, Mike Biven, Justin Waugh, Kevin Huo
DataDM is a chatbot interface the place customers speak to an AI assistant that writes code, which is then executed to reply information questions. Customers can ask questions on information processing, function engineering, information cleansing, query answering, visualizations, and even some information science modeling. They will convey their very own CSVs (that are native and personal) or can simply discover CSVs through the GitHub API, each of that are straightforward to do with a single click on add button.
Second place: Most cancers-Rx-Approve, write letters for most cancers drug denials.
Mitchell constructed Most cancers-RX, a cloud-based internet software that clinicians can use to generate letters to ship to insurance coverage corporations when requests for chemotherapy are denied. The LLM synthesizes data from the sufferers’ medical data and up to date drug analysis to create a compelling letter. These letters would each save clinicians beneficial time and cut back the time sufferers want to attend to obtain their life-saving most cancers care.
Third place: Combat Well being Insurance coverage, Routinely generated medical insurance appeals.
Holden developed Combat Well being Insurance coverage, which generates medical insurance appeals to decrease the barrier to entry for people who need to attraction insurance coverage declare denials. Combat Well being Insurances combines the output of a number of fashions (dolly-v2-12b, BioGPT, and so on.) with California’s public insurance coverage information to generate a (low high quality) artificial dataset for fine-tuning Falcon 7B onwards to generate medical insurance appeals.
Honorable mentions
Information + AI Summit 2023 Hackathon Winners
Kicking off in-person, Information + AI Summit 2023 attendees joined in on the enjoyable by attending an all-day hackathon centered round LLMs. Contributors had six hours to submit their tasks and will work in a crew of as much as 4 folks. Shoutout to MLH and the crew, Alba King, Fiona Whittington and Paul Horton, for ensuring the attendees had a enjoyable and significant expertise.
First place: SchemaSpy, Navigating Database Schemas with AI Help.
by Sai Pradeep Peri, Cameron Hutchison, Khushboo Breja, and Rik Bauwens
Second place:Â Pharma Help, Serving to folks with situations that require treatment create, keep and keep on schedule for taking their treatment.
by Joshua Bowron, Richard Ryman, Himanshu Grover, and Max Mogenis
Third place: Commit-With-A-Bang, an AI-Powered Pull Request (PR) engine for easy and insightful code administration.
by Dennis Aguillon, Patrick Younger, Priya Prakshal, and Prakshal Jain
Big congratulations to those creators! Thanks for uplifting us to proceed to maintain constructing and rising collectively, and for showcasing the highly effective influence LLMs can have throughout use instances.
Need to study extra about utilizing and constructing your personal Massive Language Fashions? Try Databricks’ expert-led EdEx programs!
