Spaces:
Running
Running
| title: Tombstone | |
| emoji: "⚡" | |
| colorFrom: gray | |
| colorTo: gray | |
| sdk: gradio | |
| app_file: app.py | |
| license: mit | |
| short_description: You're already carving it. Look down. | |
| tags: | |
| - build-small-hackathon | |
| - track:backyard | |
| - achievement:offbrand | |
| - achievement:fieldnotes | |
| - achievement:sharing | |
| - gradio | |
| # Tombstone -- You're Already Carving It. Look Down. | |
| You wrote a joke on your tombstone. Turns out it takes a lifetime to earn. This app takes what you want on the stone and works backward to show you the cost. Then it asks how you actually spend your days -- and shows you the gap. | |
| ## What it does | |
| A two-step gut check. Step one: write what you want on your tombstone -- your epitaph, the sentence you'd leave behind. The app reverse-engineers what kind of life actually earns those words. What you'd sacrifice. What you'd start this week. What you'd have to give up that you don't want to give up. Step two: describe how you really spend your days. The app puts your epitaph next to your routine and shows you the distance between the tombstone you want and the one you're currently carving. | |
| ## How it works | |
| The app sends your epitaph to Qwen/Qwen2.5-7B-Instruct via the HuggingFace Inference API with two custom system prompts -- one for each step. The step one prompt tells the model to reverse-engineer the life behind the epitaph. The step two prompt takes both your epitaph and your routine and shows the gap between them. Temperature is set to 0.8 -- warm enough to feel human, controlled enough to stay honest. When the model is unavailable, handwritten fallback responses take over and the app keeps working. | |
| ## The fallback system | |
| 8 handwritten epitaph responses and 2 handwritten comparison responses that work with zero API calls. Each epitaph fallback is matched to a specific input -- type "She was kind" and get the response about kindness being a practice not a performance, not a random one about wifi passwords. | |
| The fallbacks aren't filler. They're the standard the model has to match. Lines like: | |
| "Kindness isn't niceness -- niceness is a performance, kindness is a practice." | |
| "'He tried' sounds like a consolation prize but it's actually the only honest epitaph on the list. Everyone else is claiming results. You're claiming the attempt." | |
| "'He showed up' means being there when it's inconvenient, uncomfortable, and unrewarded. The funeral on a Tuesday. The friend's move on a Saturday." | |
| "You wrote a joke but here's what it actually takes: being the kind of person whose house other people want to be in." | |
| "Four words and a lifetime of Tuesdays. A good dad isn't the one at the big moments -- the games, the graduations. It's the one at the invisible moments." | |
| Every fallback includes one thing you'd have to give up and one thing to start this week. The model generates fresh responses when available, but the handwritten fallbacks ARE the product. They work offline, they work without credits, they work every time. | |
| When no fallback matches the user's specific epitaph, a generic response fires that still uses their exact words: "The version of you that earns [their epitaph] makes different choices than the version of you right now. Not bigger choices. Different ones." | |
| ## Why this matters | |
| Everyone has a vague idea of how they want to be remembered. Almost nobody has reverse-engineered what it actually costs. The gap between "I want to be kind" and "I spend my evenings arguing with strangers online" is measurable. Not in years. In daily decisions. Tombstone doesn't moralize. It doesn't say "you should." It just holds the epitaph up next to the routine and lets you see the distance for yourself. | |
| ## Try it | |
| - **"She was kind"** -- Two words. Hardest epitaph to earn. The app shows you what kindness actually costs when you're tired. | |
| - **"He showed up"** -- Sounds simple. The app breaks down what it means to be the person who's there every time, not just when it's easy. | |
| - **"The wifi password is on the fridge"** -- A joke that turns into a life philosophy about open doors and full kitchens. | |
| - **"He was a good dad"** -- Four words and a lifetime of invisible Tuesdays. The app shows you the moments that count and why they're the boring ones. | |
| - **"She told the truth"** -- The app shows you what truth-telling actually costs. Spoiler: it gets people fired, dumped, and disinvited. | |
| ## Built with | |
| - Gradio | |
| - Qwen2.5-7B-Instruct via HuggingFace Inference API | |
| - Python | |
| - Zero external dependencies beyond gradio and huggingface_hub | |
| ## Process notes | |
| The two-step structure came from a simple observation: everyone knows what they want on their tombstone, but nobody connects it to Tuesday. Step one is the dream. Step two is the audit. The prompts were engineered to avoid preachiness -- the system prompt literally says "don't say 'you should'" and "talk like a friend who takes you seriously when you say something you meant as a joke." That voice took several iterations to land. Too stern and it reads like a self-help book. Too soft and it doesn't stick. | |
| The fallback-first design means every hardwritten response was crafted before the model was ever called. The epitaph for "He tried his best" went through multiple drafts to find the line between honesty and cruelty -- "this sounds humble but it's actually the bravest thing on the list" sets the tone without flattering the user. The comparison fallbacks are deliberately generic but still useful, because the gap between any epitaph and any routine is usually the same shape: you know where you want the chisel pointed, you're just not looking at the stone. | |
| The example buttons in the UI are intentional. Six pre-written epitaphs plus a "surprise me" button that pulls from a pool of 16. The pool includes jokes ("Finally got some rest," "I told you I was sick") next to serious ones ("She left it better than she found it") because the whole point is that jokes have truth in them. The prompt says so: "Take their epitaph seriously, even if it's funny. There's truth in humor." | |
| Light theme. Gray gradient. Space Mono headers. The visual design is deliberately quiet -- stone-colored, not flashy. The tombstone is the loudest thing in the room. | |
| ## Demo & Social | |
| - [Demo Video](https://youtu.be/uMaNFWWLmHE) | |
| - [Social Post](https://x.com/Wayfindersix/status/2066526363473908216) | |
| --- | |
| Heuremen -- Let's stay connected and keep information free. | |