- Conversational AI assistant for email and messages
- A tool for companies to connect task management systems
- A service that buys things on your behalf at the lowest prices
- Custom Telegram client for crypto speculation
- Custom Telegram client for enterprise collaboration
- Competitor monitoring as a service
- Lightweight offline Notion app
- Conversational assistant for toddlers
- Programmatic endpoint comparison tool
- Syncing data from any internal tool
- Problems that pique my curiosity
- Native ad embedding in LLM responses
- SDK to manage browser ad targeting APIs
- O-1 Visa as a service
Conversational AI assistant for email and messages
Problem: Professionals are too busy to sit down in front of a phone or computer to triage email.
Solution: Phone-based service where an AI summarizes each email and lets you act on emails in a conversational audio manner. This service can be extended to messages, to-do lists, etc.
Walkthrough: Connect your email on a website. After that, call a number and enter in a verification code that is texted to you. An AI will summarize each email and let you reply with natural language to take action:
- AI: “Ikon pass says season passes are now available, and a 20% early bird discount is available until tomorrow.” You: “Great. Add to my to-do list tonight to buy one. Archive this email.”
- AI: “Kevin and Jack accepted your calendar invite for a Project X sync tomorrow. Your Amazon order for a fan has shipped with a delivery date of Wednesday. I’ll archive these.”
- AI: “Doordash sent you an FSA receipt for your children’s ibuprofen delivery.” You: “Tag it as
- AI: “You got three recruiting emails yesterday from AtoB, Runway, and Doordash.” You: “Can you tell me about AtoB?” AI: “AtoB is building fintech for commercial fleets, are backed by General Catalyst & Y-Combinator, and have raised $100M as a Series-B startup.” You: “Thanks, not interested in these. Archive please.”
- + Other examples to allow you to unsubscribe, snooze, emails etc.
My experience: I’m a productivity geek and a dad. I spend an hour a day washing baby bottles and pump parts, and spend most of that listening to podcasts or zoning out to music. By the time my daughter goes to sleep, I’m too tired to handle my emails, messages, or digital to-do list, and they keep piling up. I’d love a way to turn that hour a day into productive time. For others, this may be time spent doing laundry or in a commute. I’m not rich enough to afford a dedicated virtual assistant, and don’t have enough tasks to merit one either.
A tool for companies to connect task management systems
Problem: When engineering teams for data providers and customers work together, EM/PMs spend repetitive and manual effort relaying information over Slack and Google Docs.
Solution: Tool for data providers to sync task status from their internal task management system to a shared page that customer companies can access.
Walkthrough: A data provider works with many customers. These customers will ask for custom features or file bugs, each of which is logged into the data provider’s task management system. The data provider connects a tool to their Jira system, and selects which Jira tasks are related to each customer company. As the task progresses or a new ETA is estimated, the shared page updates and a notification is sent to a Slack channel.
In the future, a data provider’s eng team can use this tool to share internal roadmaps with date estimates or relevant service degradations. Customers are able to directly file bugs into a data provider’s task management system or page their on-call engineers.
My experience: When we had asks for a data provider, we filed bugs by writing in a shared Slack channel (later Google doc), and tagged the other team’s PM. That PM would ask their eng team for an estimate, then share it back to me. I had to set a weekly reminder to ask for an update. Sometimes, they completed an ask ahead of time but forgot to notify us.
BD team has said that almost all crypto partnerships are via Telegram, which was not built for enterprise collaboration. It’s tough to schedule time, get approvals, or have a common workspace.
A service that buys things on your behalf at the lowest prices
Problem: When I make a large purchase, I want to get the biggest savings, but online tools quickly apply discount codes, but never discounted gift cards.
Solution: Chrome extension that shows you how much you can save with a gift card, and lets you easily purchase a gift card for use.
Custom Telegram client for crypto speculation
Problem: Users using Telegram trading bots find it overwhelming with a new lengthy text message with every click.
Solution: A custom Telegram client that makes interacting with trading bots feel like using an app, where buttons stay in one place and information is shown in a streamlined manner.
Walkthrough: A user uses our custom Telegram client with Unibot. Buttons and menu options stay in one place, unchanged information like links to website and tutorial is hidden, and wallet information and balances is visually represented instead of using ASCII characters. There are two extra tabs at the bottom - one is a global feed of transactions and access to a crypto bot app store, another is a portfolio page showing P&L over time as well as positions held.
My experience: Telegram trading bots, like Unibot, are increasingly how users trade crypto. Telegram itself is leaning into this by launching a self-custodial crypto wallet. However, Telegram was built as a messaging service, not as an app.
Notes and Trends: Telegram is built on open source software. Account abstraction for Ethereum is gaining steam, and most new retail users will interact with crypto via apps like Telegram that minimize complexity (signatures, gas, etc) while offering composability (take your wallet to another service).
Custom Telegram client for enterprise collaboration
Problem: Most crypto companies handle partnerships over Telegram, but this is an ineffective channel to manage tasks, scheduling, and message search.
Solution: A custom Telegram client that integrates with Slack, CRM, Jira, and more.
Walkthrough: Business development connects with another crypto team to schedule Twitter spaces or joint promotions. They use Telegram, and use a custom Telegram desktop client, which allows them to easily track asks in a CRM or task management system so it doesn’t slip through the cracks.
My experience: Crypto companies increasingly use Telegram, given the pseudonymity of many founding teams. For enterprise uses, it’s easy to use Slack’s built-in “Save for later” functionality, plus integrations like filing tickets directly into CRM and task management solutions.
Competitor monitoring as a service
Problem: It’s hard for product and design teams to monitor feature updates from competitors.
Solution: A service that continually refreshes and monitors apps and webapps for feature changes.
Walkthrough: For a company with a prominent mobile app, they’ll identify a set of core pages that a competitor has. Every app update, this service navigate throughout their app, takes screenshots of each page, and stores them in Figma. We’ll call out new toggles or features, while not flagging changes due to live data (eg. prices). We’ll also document the entire app flow in an easy-to-use flowchart.
My experience: At two separate companies that had mobile apps, we have a channel where we posted screenshots of competitor apps when we notice a change. On the product, design, and eng teams, we also had issues understanding what’s the latest design for our apps. This is because specs and designs get out of date, especially after bug fixes, and the screen in question is in the middle of a process (eg. mid-ride, mid-transaction).
Lightweight offline Notion app
Problem: Offline mode has been the #1 user request for Notion for many years, but Notion will likely never implement this because too many features are handled server-side.
Solution: Build an offline Notion app with limited functionality.
Walkthrough: I download an offline Notion app, which prompts me to grant access to my Notion via their API. Depending on the number of pages I have, it’ll download all pages, or just my favorites, to my device. Viewing and editing notes will work, but advanced functionality like database operations (which are currently handled server-side) will not.
My experience: I’m a Notion fanatic and store everything in Notion. I want to see my grocery list, frequent flyer numbers and flight info, and other reference information, especially in places when I don’t have, or have spotty, mobile data, like large supermarkets or airports. I’m a power user with thousands of notes, and dozens of complex databases and automation, so recognize that Notion can’t store these locally.
Conversational assistant for toddlers
Problem: Toddlers watching YouTube end up watching and listening, but not talking, because they understand interactivity.
Solution: Build an audio assistant that converses with toddlers and helps them practice words they’re not saying well. This could be via an iPad app or TV attachment.
Walkthrough: An iPad app that provides toddler programming, stopping intermittently to ask them to repeat a word. If they don’t repeat it, to try a few more times. If they do, to provide a congratulatory experience.
My experience: We let our daughter watch toddler programming when we need a moment, and it’s often Ms. Rachel. While she’s fixated on the TV, she doesn’t talk as much as she would if we sang or spoke with her on a video call.
Companies in this space, but I want this product ASAP:
- PBS Kids is exploring AI-powered interactive TV shows where kids can reply
- Hellosaurus add some interactivity, but it’s not audio-based
Programmatic endpoint comparison tool
Problem: Programmatic comparison between two endpoint providers isn’t easy to do.
Solution: A tool that given two API endpoints and params, identifies differences in response data and latency, and makes it easy to dig further.
Walkthrough: I’m considering whether to migrate to a new data provider who gave me a test endpoint. I fetch a bulk number of requests to both my current and this new test endpoint, and specify the response fields I care about as well as any special behavior (eg. a 1% difference is fine for a pricing field, or that I do/don’t care about the # of digits returned). It shows the latency distribution for this new API provider, which has a lower median latency, but for 2 cases, has much higher latency. Also, response field A had 3 responses that differed from our current provider, and it makes it easy to dig further in by showing each data point side by side and letting me view context inline.
My experience: When evaluating different data providers, it’s hard to compare latency and data. For latency, we often only have telemetry in our apps, so we’d have to make a build pointing to their endpoint, and simulate a bunch of requests. To save work, we often just ask providers themselves for their own measurements but their test cases may be different from ours. For data differences, programmatic diff-ing requires writing one-off scripts, so we often just do a bug bash of our most important cases instead, but this misses the long-tail.
Syncing data from any internal tool
Problem: It’s manual and repetitive copying results from internal tools like custom dashboards or A/B result scorecards into docs.
Solution: An add-on for Google Docs or Notion to sync data live from another browser tab.
Walkthrough: I open up an internal dashboard that is behind my company OAuth, and a Google doc that our team uses for meeting notes. In Google docs, I use an add-on to insert an “object” onto the page. This object specifies which fields in the dashboard I want to display in Google docs, and pulls it live data from my dashboard tab. Unclear about implementation and security, but hoping that this can be done locally on-device with a Chrome extension (which can view content on all my tabs) and a Google Docs add-on (to allow for custom content type).
My experience: On a weekly basis, we would review ongoing experiments and data on different market regions. It required opening an internal page, and copy and pasting data points one by one into a Google Doc.
Problems that pique my curiosity
Here are ones that I haven’t personally encountered and do not have domain expertise in.
Native ad embedding in LLM responses
Problem: Costs of training and running LLMs are very high such that best models aren't deployed and aren’t free
Solution: An API to embed ads natively into text, photos, or videos. This is done in realtime, based on the content returned, contextual situation, and user details. A smaller LLM could identify opportunities and regenerate the response to include the ad.
SDK to manage browser ad targeting APIs
Problem: As cookies are deprecated, every browser has a different API to share user cohort and retargeting data
Solution: An easy-to-use SDK for publishers that seamlessly works with each browser’s ad targeting APIs to maximize revenue per website visitor.
Walkthrough: A publisher adds a few lines of code to their website, which passes along the API responses that the browser provides about the user to the publisher’s supply-side platform. Without this SDK, a publisher would need to manually implement code to interpret API responses from each browser, or if they don’t, risk losing revenue because that browser’s users will be less targetable.
Details: Cookies are being deprecated in Chrome in early 2024. Chrome’s replacement is a set of APIs called Privacy Sandbox, while Firefox is working on Interoperable Private Advertising, and Safari has already implemented Private Click Measurement. The long-tail of browsers like Brave, Opera, UC Browser, etc. also each have a different privacy implementation.
O-1 Visa as a service
Problem: O-1 Visas are the best way for founders to work in the USA, but they’re difficult to get, the process is unclear, and are very expensive. Tweet
Solution: A tech-enabled platform to make it easier, and cheaper, to get them
Thoughts: O-1 is a lot less paperwork, but getting peer endorsements, so tech helps a lot less. It’s also a feature for Boundless.com, rather than a new product.