Startup and Project Ideas

After years of building software professionally, I have dozens of ideas I’ve collected. Realizing I won’t ever complete most of them, I’ve decided to give them away. I’m starting with 20 at random.

1. Styled eyes for Apple VR headsets. Virtual adornment of your face image seems a lot of fun.
2. Fractal mirror. A device that captures your image from various angles and displays them back to you. Point to parts of your body to track rashes, moles, muscle growth, or whatever you want. It can show close-ups of certain areas like the back of your head, or eyes and eyebrows for faster and more accurate grooming. An augmented interface could give health and style tips.
3. Inverse app framework. Most web frameworks don’t include logging, metrics, and library extensions. There’s value in making these into supplementary standard libraries. Once there’s standardization, adapters can be made for the popular frameworks. Currently, there’s no way to log or monitor installed dependencies in applications, but if an “inverse framework” were widely standardized as a construction parameter by dependencies, they could instrument against the same interfaces, giving the application total control.
4. Capsule Hotel Busses or Planes. Extremely low-cost travel. Revolving sushi-bar-style food and merch could make it somewhat fun and comfortable.
5. AI topic tracker. Build an engine that catalogs user intents as they use an application (either implied or explicitly given), and finds topics to research, questions to answer, and content items to organize. This might include learning areas, learning next steps, or keeping track of subtopics as parent topics are investigated. Use this to prompt new IA and inline conversations.
6. TODO AI. Like TODO MVC and Realworld help developers compare stacks given a common set of requirements, TODO AI compares AI agent architectures. TODO AI requirements are additive to the TODO MVC ones—requiring your TODO list to be automatically processed by one or more agents as you add items. A custom evaluation framework would score TODO AI task effectiveness. An interesting concept here would be a prompt to shape how TODOs are prioritized, which completion criteria are required in the abstract, and so on.
7. Photography alarm. Crowdsource schedules for various things worth photographing and find schedule combinations that make for interesting photographic opportunities. This could include celestial schedules, festivals, anniversary events, air shows, construction projects, sports events, and anything worth photographing or photographing together. Alert photographers when events and event combinations are coming up, where they should travel, and how they should prepare their equipment.
8. Wormhole-based CRDTs. This idea combines Magic Wormhole and conflict-free replicated data types, enabling an application paradigm based on the pulling and syncing of content in a decentralized way. You might think of this as Twitter for data but without the need for a centralized broker. You periodically connect to people whose data you follow via wormholes, then automatically sync content they maintain. One basic example is sharing a structured description of secrets on a dev team.
9. AI tiled layouts. It should be possible to open and close applications and position their windows with AI descriptions. Workstations should respond to user needs, quickly arranging huge numbers of windows and applications needed to do work. This could be done with code generation on top of i3, but a declarative interface for describing desktops would be easier for the AI. A frequently-changing periphery would be a powerful feature.
10. AI CI/CD. Imagine a CI pipeline that was sensitive to a team’s current workload and morale, and could support them through technical challenges of varying difficulty. Such a system could strategically decrease testing demand during periods of creative exploration, possibly being more strict with the stable, or security-sensitive interfaces. If it noticed a decrease in quality in some area (ie. stories without acceptance criteria), it could flag work as incomplete. Designed well, this could include more of the organization in the development process by flagging problems further upstream.
11. API babel. Send the output of one API to the inputs of another with an LLM-based glue layer that automatically fixes the semantic and structural misalignment between APIs. A big missing piece of the web platform is that you can’t log out from all your services, change your home address from a centralized place, set your social media status or avatar, or perform administration broadly. This would be a first step towards fixing those problems.
12. Digital security patrol. The idea here is to have site crawlers look for violations of terms in common digital platforms. This might be outright illegal activity based on text in a post, or could involve image analysis like verifying an apartment building prohibiting short-term rentals isn’t listed on any platforms.
13. GTD scheduler. One of the bigger issues with the Getting Things Done (GTD) productivity theory is that there are too many processes to remember and plan for. An application could prompt for your input at these daily, weekly, and monthly intervals. AI could privately triage work and personal tasks, and mine project repositories and inbox items for logical next actions.
14. Personal ranker for public officials. Search legislative voting records or legal decisions by public officials, focusing either on single issues or broad-platform comparisons, where the user can use this data to like and dislike these representatives. Help people decide which regions have values-aligned representatives, which institutions are funded for those issues, and which representatives support them. Help people find alternative candidates when incumbents aren’t performing well.
15. Headless spreadsheet. Using a technique in the style of Kent Dodds’ Downshift library, build a headlesss frontend component similar to Pandas data frames but for reactive dataflow. The main benefit is removing the UI complexity associated with a full spreadsheet, instead requiring users to skin separately or extract data via snapshots and use with an ordinary table. Flexibility could be added that supports novel spreadsheet experiences or non-tabular reactive experiences like data visualizations backed by reactive tables.
16. Dynamic information architecture components. Build menus that add deep or custom items based on user intent. Filtering on list views could be pre-applied. The default query of a data visualization could be intent-based. Applications could respond to intents inferred based on how users are clicking and searching and also could react as new server data arrives.
17. Code as art. Code can be beautiful, so why not make it into art? Given a piece of code, generate supporting visuals and style the code to highlight the interesting parts. A website could be built around this to support social commentary, personal collections, and voting. A store could exist for printing posters, custom books, and other media.
18. AI-augmented Code Libraries. Fork popular libraries and give them targeted LLM support. For example npm install ai-underscore ai-create-react-app.
19. Emoji-based Trip Phrases. Instead of limiting tripphrases to letters and words, use more interesting base symbol sets.
20. Personal scoring of hexagon maps. A person is free to use services and experiences from places across the globe, but most people don’t have a good way to operate on a global level. You might do your banking, buy your furniture, get dental work done, go surfing, and take courses all in different countries. I’d like a mapping application that crowdsourced global data like Pieter Level’s apps do but allows a person to build heatmaps with polynomial scoring based on a very long list of survey questions. A system like this could match and rank regions to live and spend time. The answers to questions would rank regions of the world and tell you what kinds of experiences to seek there.