AI Hackathon Stack

March 12, 2024 · View on GitHub

⚠️ This is a very new/immature list, created for the Latent Space Demo Day. The category labels not that well thought through. Please get in touch if you have better ideas for how to organize this, it is welcome.

Below is a curated list of useful tools and examples for developers tackling AI Hackathons, together with useful hackathon-centric offers and templates for you to get started quickly.

If you represent a tool or vendor, please feel free to send in a PR for your tool, but note that we may reword or reject your submission based on subjective fit.

LLM/Prompt Engineering Apps

Templates/Example Apps/Tutorials
- Browse through all the HuggingFace Spaces
- Deploying AI Applications on Vercel Overview and open source templates
  - with streaming edge functions
- a16z demo app https://github.com/a16z-infra/ai-getting-started/
  - a16z AI companion starter kit
- LangChain + Pinecone GPT assistant tutorial, from PineCone
- https://github.com/nat/natbot: Drive a browser with GPT-3. Demo tweet video
- https://github.com/menloparklab/langchain-cohere-qdrant-doc-retrieval This Flask backend API takes a document in multiple formats (.txt, .docx, .pptx, .jpg, .png, .eml, .html, and .pdf) and allows you to perform a semantic search in 100+ languages supported by Cohere Multilingual API. Qdrant vector database is used to save embeddings. tweet
- ChatGPT Plugins
  - https://replit.com/@bardia/ChatGPT-Todo-Plugin
  - https://replit.com/@bardia/ChatGPT-Retrieval-Plugin
  - https://github.com/transitive-bullshit/chatgpt-plugin-ts
  - https://supabase.com/blog/building-chatgpt-plugins-template
  - Build a ChatGPT Plugin on Cloudflare Workers: Tutorial; Quickstart
- Gemini/GPT4 Vision/Speech starter
Tools to explore
- LLM APIs
  - OpenAI needs no introduction. Cookbook, Docs
  - Replicate - OSS models galore, esp image stuff. Showcase, Chat template
- Prompt Engineering Libraries
  - LangChain - Building applications with LLMs through composability. Discord. ChatGPT tutorial.
  - Lambdaprompt - Build, compose and call templated LLM prompts
  - https://github.com/ianarawjo/ChainForge An open-source visual programming environment for LLM experimentation and prompt evaluation.
- Embeddings
  - Chroma: Chroma is the open-source embedding database. Chroma makes it easy to build LLM apps by making knowledge, facts, and skills pluggable for LLMs. Docs, Discord, @ the founders
  - Pinecone: The Pinecone vector database makes it easy to build high-performance vector search applications. Docs, Events/Forum/Showcase.
  - Also Weaviate, Milvus/Zilliz, Qdrant, Vespa, pgvector. Seeking: comparison articles between these options. nice chart here
- have something to add? send a PR!
Hackathon Entry Examples
- Automatic permit application generation for climate tech companies & carbon dioxide removal (tweet)
- a personalized learning curriculum generator (tweet)

Code AI tools

Where the focus is building AI tools for other developers.

Tutorials
- Build a GitHub support bot with GPT3, LangChain, and Python - Dagster blog ChatGPT clone (tweet thread). Led to Astro's HoustonAI: Astro AI support Bot
Tools to explore
- OpenAI Codex needs no introduction.
- Copilot Explorer Tools for those reverse engineering copilot. You can also peek at GitHub Copilot Labs.
- https://github.com/fauxpilot/fauxpilot This is an attempt to build a locally hosted version of GitHub Copilot. It uses the SalesForce CodeGen models inside of NVIDIA's Triton Inference Server with the FasterTransformer backend.
- The Stack: 3 TB of permissively licensed source code in 30 programming languages https://huggingface.co/datasets/bigcode/the-stack
- have something to add? send a PR!
Hackathon Entry Examples aiming for Open Source ones, for inspo and study
- 🏆 GPT is all you need for backend: a backend and database that is entirely LLM-powered. (tweet)
- KnowledgeGPT
- Chatbase.co: LangChain + Pinecone + OpenAI
- GPT-3 Auditor: scanning code for vulnerabilities with LLMs. https://github.com/handrew/gpt3-auditor
- Gptcommit: Never write a commit message again (with the help of GPT-3)
- santacoder typosaurus - semantic linter that spots errors in code
- stackoverflow.gg https://twitter.com/bentossell/status/1622513022781587456
- Buildt - AI-powered search allows you to find code by searching for what it does, not just what it is. https://twitter.com/AlistairPullen/status/1611011712345317378

Audio/Visual/Multimodal Apps

Tools to explore
- https://www.synthesia.io/ AI video and voice creation
- Hosted Stable Diffusion
  - https://replicate.com/stability-ai/stable-diffusion
  - https://replicate.com/cjwbw/stable-diffusion-v2
- Stable Diffusion UIs
- Transcription/Whisper
- Multimodal, semantic video search
  - Twelve Labs: Example: https://youtu.be/484hydNEJC0
    - Docs https://docs.twelvelabs.io/docs/understand
- Youtube semantic video search
  - Atlas
    - Demo: https://twitter.com/tomiwa1a/status/1617373731135029249
    - How it works: https://atila.ca/blog/tomiwa/atlas
    - Source Code: https://github.com/atilatech/atila-core-service/tree/master/atlas
- Computer Vision
  - Roboflow: Build classification, object detection, and instance segmentation datasets and models. Explore over 10,000 open source models and millions of labeled images on Universe. (See Roboflow Hackathon projects like visual chess solving apps, automatic dog treat release.)
  - 10k open source models, 10M+ open source labeled images: https://universe.roboflow.com
  - LlavaVision: Bakklava + Llama.cpp
  - open source macos copilot: https://github.com/elfvingralf/macOSpilot-ai-assistant
    - - Use a keyboard shortcut to take a screenshot of your active macOS window and start recording the microphone.
    - Speak your question, then press the keyboard shortcut again to send your question + screenshot off to OpenAI Vision
    - The Vision response is presented in-context/overlayed over the active window, and spoken to you as audio.
    - The app keeps running in the background, only taking a screenshot/listening when activated by keyboard shortcut.
    - I's built with NodeJS/Electron, and uses OpenAI Whisper, Vision and TTS APIs under the hood (BYO API key).
- have something to add? send a PR!
Hackathon Entry Examples
- HouseGPT generates raw MIDI data directly from few-shot prompted GPT-3 to create 🎶 house music 🎶 🔊 (tweet)
- Rap Battle - Pick any two people and it will generate a rap battle on the fly, using GPT-3 for lyrics, wavenet for vocals, and stable diffusion for the avatars.
- Game of Life, where each alive cell is a whimsical happy Stable Diffusion image and each dead cell is an eerie, dark Stable Diffusion image, all of which evolve over time. (tweet)
- NotesGPT: generate action items from your notes in seconds. Powered by Convex, Together.ai, and Whisper.

AI Infra and Tooling

This category is for infra and tools catering to AI app developers, in contrast to apps (below) which have other kinds of end users in mind.

Tools to explore
- Serverless GPUs
  - https://lambdalabs.com/ GPU cloud built for deep learning. Instant access to the best prices for cloud GPUs on the market. No commitments or negotiations required. Save over 73% vs AWS, Azure, and GCP. Configured for deep learning with PyTorch®, TensorFlow, Jupyter
  - https://www.banana.dev/ Scale your machine learning inference and training on serverless GPUs.
  - Cerebrium Deploy to serverless CPUs/GPUs with ~2 second cold start times.
  - https://brev.dev/ The simplest way to create a dev environment with a GPU. Don't worry about dependencies, CUDA, SSH, or anything else. Up to 94% cheaper GPUs than AWS. For example: here's a guide to get a GPU dev environment to train your own Dreambooth model in ~4 min 🤙 https://brev.dev/docs/guides/dreambooth
  - https://www.coreweave.com/
  - vast.ai (strong rec)
  - runpod, fluidstack, potentialy sfcompute.org
  - https://www.paperspace.com/ Paperspace provides tooling for training and serving, with serverless GPUs, development environments, and workflow management systems.
  - More serverless discussions: https://news.ycombinator.com/item?id=34742087
  - Seeking: hackathon-relevant examples and tutorials for each of these tools
- Model Serving
  - https://www.baseten.co/ Serverless backend for building ML-powered applications. Build apps with auto-scaling, GPU access, CRON jobs, and serverless functions.
  - https://replicate.com/ Run models in the cloud at scale.
  - https://www.cerebrium.ai AWS Sagemaker alternative. Deploy to serverless CPUs/GPUs with ~2 second cold start times. 40%+ cost saving.
  - (emerging) https://docs.preemo.io/docs host oss models
  - https://modal.com run or deploy machine learning models, massively parallel compute jobs, task queues, web apps, and much more, without your own infrastructure.. Example serving Stable Diffusion API, BERT in 34 lines of code
  - https://www.steamship.com/ Managed Backend for AI services LangChain example
  - Cloudflare Constellation - deploy models and run inferencing on Cloudflare's edge
  - Seeking: hackathon-relevant examples and tutorials for each of these tools
  - Seeking: tutorials for serving Whisper, other LLMs
- Fine-tuning
  - Blueprint by Baseten Fine-tuning and serving of foundation models. Docs. In alpha; reach out on Discord to Amir H. for help.
  - Cerebrium Fine-tune and deploy foundational models or bring your own models. Serverless ML Infrastructure. Docs
- Evaluation
  - Zeno Interactive data management and evaluation for foundational models. Docs and Demos.
- have something to add? send a PR!
Hackathon Entry Examples
- a key-value store to enable long-term memory in language model conversations (tweet)

Misc Useful Resources

Looking for tutorials, writeups, guides, inspo, requests for startups.

Hackathon Recaps for Inspiration
- Scale AI Hackathon https://scale.com/blog/generative-ai-hackathon
- AssemblyAI Hackathon https://twitter.com/AssemblyAI/status/1602717569659682816
have something to add? send a PR!