Open Source AI Playground for Prompt Engineers
Hi HN, We're a small team building AI tutors out of India, and as you might guess, this means we spend a ton of time writing, testing, and refining prompts for LLMs. When we started out, we were using the OpenAI playground but things became tedious when we wanted to compare responses from different models. We tried a bunch of other playgrounds but found them lacking in some features so we built our own. Quick Links: Github: https://github.com/supernova-app/ai-playground Hosted demo: http://playground.getsupernova.ai Demo video:…
What it does
In the maker’s words, at launch
Hi HN, We're a small team building AI tutors out of India, and as you might guess, this means we spend a ton of time writing, testing, and refining prompts for LLMs. When we started out, we were using the OpenAI playground but things became tedious when we wanted to compare responses from different models. We tried a bunch of other playgrounds but found them lacking in some features so we built our own. Quick Links: Github: https://github.com/supernova-app/ai-playground Hosted demo: http://playground.getsupernova.ai Demo video: https://www.youtube.com/watch?v=I01_t75FT-c TLDR: Main features are: - Monaco editor for writing prompts. - Variable support in prompts {{}}. - Syntax highlighting for tags like XML. - Generate multiple completions with same model. - Chat with multiple models simultaneously. - Save prompt and conversations as JSON. - Easy to self host. Key Features: 1. Monaco Editor for Writing Prompts When we were working on long, detailed prompts, writing them in plain text felt clunky and error-prone. Small issues—like missing a tag or having weird formatting—could break things. So, we integrated the Monaco editor (used in VS Code). It gives us: - Line numbers (so we don't get lost in long prompts). - White space detection to catch formatting issues early. - Syntax highlighting for tags like XML. - Code folding to collapse parts of a prompt we're not actively working on. These might sound like small things, but they've been a huge help when we're dealing with large, complex prompts that need constant tweaking. 2. Variable Support for Dynamic Prompts You can define placeholders in your prompt using double curly braces ({{ }}) and fill them in via a friendly UI. 3. Testing for Consistency Across Completions One of the hardest parts of building AI tutors has been ensuring reliable outputs. Even when a prompt seems fine, it can fail unexpectedly—or worse, it works sometimes but not always. To address this, we made it easy to generate multiple completions from the same model at once. This lets us quickly see: - If the prompt is consistently producing good results. - Where the AI might misinterpret our intent. For example, we'd often run 5–6 completions to see if the AI consistently understood our instructions, rather than getting lucky once or twice. 4. Comparing Models Side by Side The main reason we built the playground in the first place. You can set up API keys for multiple providers and see how different models handle the same task. This helped us: - Optimize prompts for specific models. - Choose the best model for a particular use case. 5. Saving Conversations as Test Cases Another pain point was testing how a prompt or conversation would evolve over time. Sometimes, we'd go back and forth with the AI to simulate real user interactions, but we had no way to save that conversation for future reference. Now, with the playground, we can save these conversations as test cases. Here's how it works: 1. We create a conversation (or simulate a long interaction). 2. At any point, we can save it as a JSON file. 3. The JSON includes the full conversation, the system prompt, and any variables we used. We then use this JSON file and use it in our code for running test cases or run evals. 6. Simple Self-Hosting Finally, we wanted to make sure the playground was easy for others to set up. The only dependencies are: - A Postgres database. - API keys for the AI providers you want to use. It supports Google login, so you can setup OAuth and can restrict access to only your domain. The app is open source and we are running a hosted version of it here: http://playground.getsupernova.ai. You can check out the repo here: https://github.com/supernova-app/ai-playground. It's easy to self-host, and we're actively working on new features. If you give it a try, let us know what you think! Feedback, feature ideas, and contributions are all welcome.
Does the same job
all alternatives →
- IMI'm 13 and built an OpenAI o1-like AI model – would love your feedback2024 · ai.pixelverse.tech · ▲6
I’m Hayden, a 13-year-old developer based in Australia, and I’ve built a CoT logical thinking and reasoning AI model similar to OpenAI o1. It's powered by open source small models like Llama 3.1 and 3.2 and I would love for you to try it and share your feedback with me. You can try it here: https://ai.pixelverse.tech/app/cortexchat I built it just for fun and launched it a day after the o1 release. It's not perfect yet but its still amazing to see how a detailed prompt can have such a difference on the quality of the LLM response! Please let me know any feedback or…
More ai this month
the category →
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 17d ago · simedw.com
Astute▲585Automate your B2B brand going viral, with new media creators
AI · 18d ago · company-app.joinastute.com
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
AI · 26d ago · cactuscompute.com
Launched alongside, January 2025
the whole month →- IM
Hello! I'm Byran. I spent the past ~6 months engineering a laptop from scratch. It's fully open-source on GH at: https://github.com/Hello9999901/laptop
Dev tools · 2025 · byran.ee
- TITetris in a PDF▲1,289
I realized that the PDF engines of modern desktop browsers (PDFium and PDF.js) support JavaScript with enough I/O primitives to make a basic game like Tetris. It was a bit tricky to find a union of features that work in both engines, but in the end it turns out that showing/hiding annotation "fields" works well to make monochrome pixels, and keyboard input can be achieved by typing in a text input box. All in all it's quite janky but a nice reminder of how general purpose PDF scripting can be. The linked PDF is all ASCII so you can just open it in a text editor, or have a look at…
Life & fun · 2025 · th0mas.nl
Create lifelike, personalized AI avatars from text prompts
AI · 2025 · jogg.ai
