Final month, we partnered with Lablab.ai to host the Llama 2 Hackathon with Clarifai. The Hackathon aimed to nurture creativity within the growth of AI-first merchandise utilizing the highly effective Llama 2 LLM from Meta, with any one of many 7B, 13B, and 70B parameter variants. A complete of 1,765 fanatics comprising 227 groups registered, leading to a powerful 51 undertaking submissions.
With so many nice entries, the duty of choosing the winners proved to be a difficult one! Listed below are among the standout tasks.
Constructing with Llama 2 & different fashions from Clarifai Group
The problem required contributors to make Llama 2 the core part of their tasks whereas additionally leveraging different groundbreaking fashions like Steady Diffusion XL, Code Llama, GPT-4, and extra.
1st Place 🥇: SightCom
SightCom, received the highest prize with a cellular app designed for visually impaired people utilizing the Flutter framework. It makes use of Blip 2, Coloration Recognition mannequin, Optical Character Recognition mannequin, Llama 2 7B and BARCODE-QRCODE-Reader workflow within the backend.
It affords 5 distinctive accessibility options, all seamlessly activated by means of voice instructions:
1. Scene Description: Utilizing Blip 2, a picture captioning mannequin, the app narrates the objects it identifies by means of the digital camera, serving to customers perceive their environment.
2. Coloration Recognition: It identifies the colours of objects, enhancing their consciousness of the visible world utilizing Clarifai’s Coloration Recognition Mannequin
3. Textual content Recognition: Utilizing an Optical Character Recognition mannequin, textual content inside the digital camera’s view is transformed into spoken phrases.
4. Product Reader: By scanning barcodes with the BARCODE-QRCODE-Reader Workflow, the app offers detailed details about merchandise, empowering customers to make knowledgeable decisions whereas purchasing.
5. QnA Chatbot: Powered by Llama 2 7B mannequin, the app additionally affords a digital assistant that responds to customers questions.
2nd Place 🥈: FINGU: Your personal Private Finance Assistant
FINGU received second place with its answer geared toward dealing with private funds. Think about FINGU as a monetary knowledgeable that continually learns out of your monetary interactions, behaviors, and market traits.
This is how FINGU works:
1. Person Interplay: The bot interacts with customers by means of Telegram messages. It takes instructions, prompts, and uploads of CSV information as inputs.
2. Llama 2 7B Mannequin: To generate significant monetary recommendation, it makes use of the Llama 2 7B mannequin to grasp consumer inputs and generate contextually related responses.
3. Function Immediate Optimization: The bot begins every dialog with a rigorously designed function immediate that guides the bot’s conduct. This function immediate offers helpful responses, eliminating extreme randomness.
4. CSV Interplay: Customers can add CSV information, and the bot processes the content material of the file utilizing common expressions to take away date and time codecs. It then makes use of this processed content material to formulate a dialog with the consumer.
5. Reminiscence Administration: It additionally employs LangChain’s reminiscence administration that enables the bot to keep up separate dialog histories for every consumer. This permits the bot to recall previous interactions and supply extra personalised recommendation.
It additionally employs end-to-end encryption to maintain info confidential and reliable.
third Place 🥉: Slide Deck AI
SlideDeck AI, the third-place winner, is an AI-powered answer that creates presentation slide decks in just some easy steps.
Begin by describing the subject of your presentation, and based mostly in your description, SlideDeck AI generates the content material and descriptions in your slides utilizing the Llama 2 13 B Mannequin. It then transforms your pure language content material into structured JSON knowledge utilizing GPT 3.5. Subsequent it creates a PowerPoint slide deck utilizing that knowledge.
However SlideDeck AI does not cease there. It additionally fetches useful assets from the online in your presentation subject, supplying you with worthwhile analysis materials, and it even creates idea artwork associated to your subject utilizing the Steady Diffusion XL 1.0 mannequin.
Llama 2 Hackathon Finalists 🏆
IntelliSum – Clever Summarizer for Textual content
IntelliSum is a browser-compatible extension powered by Llama 2 to reinforce the online expertise. It does Summarization and extraction of textual content from a given URL, generates AI prompts utilizing the summarized textual content and may generate pictures for that immediate utilizing the Steady Diffusion mannequin.
Positively-Not-Jarvis one other finalist is an app for immediate and correct question decision. It makes use of Llama 2 at its core, understands numerous technical, informational, or informal queries and delivers swift and exact responses.
It may be built-in into any Slack workspace as a customized bot. Plus, it retains studying and bettering out of your questions, adapting to your workforce’s distinctive communication model for even higher responses over time.
DefendIQ Empowering Navy Operations by means of AI
DefendIQ is a safety answer that mixes AI and IoT gadgets for unparalleled safety. The customizable AI adapts to distinctive wants, whereas IoT gadgets like cameras and drones work collectively to verify every thing stays protected. DefendIQ enhances human capabilities by studying from real-time responses and coaching in your knowledge. It makes use of Clarifai’s weapon detection and Meta’s Llama 2 70 B fashions.
Llama2 GPT CodePilot VSCode Extension
Present it with enter, and it generates related responses, making coding experiences extra handy. You could find it on the VSCode extension market below the identify Llama2-GPT-CodePilot.
Because of Lablab.ai for partnering and internet hosting this hackathon, and congratulations to all the hackers that got here collectively to make it a hit. You’ll be able to try all the submitted ideas, prototypes and pitches within the Lablab.ai submission web page right here.
This was our first hackathon utilizing generative AI and LLMs, and the good concepts, vitality, and collaboration from all hackers concerned helped us understand this to a better degree than we may’ve hoped. We had been actually delighted and stunned by each the quantity and high quality of the submissions.
Need to keep within the loop for the following Clarifai Hackathon? Observe us on Twitter to remain updated!