Alex Atallah

Note: this post has been moved to alexatallah.com

For the 2017 TechCrunch Disrupt hackathon, Devin and I make a marketplace for sharing Internet access. Using your router, you as a host can open a special WiFi hotspot and earn cryptocurrency when people use it.

When demand is high, hosts lease unused bandwidth to clients and thereby “mine” Wificoin (WIFI). When demand is low, they verify the integrity of other hosts’ packets, using a blockchain to reach consensus. Clients can join any Wificoin hotspot to start using the Internet, paying (with fiat currency or with WIFI) per byte transferred instead of paying for every minute.

Finishing this project involves building a centralized RADIUS-compliant server to authenticate clients to the network as they hop from hotspot to hotspot. The ISPs who work with us are then rewarded by the growing value of the WIFI token on their connected routers.

Here's TechCrunch's writeup of the project:

Apriori performs statistical analysis with the tap of a finger. Daniel Posch and I built it for the Salesforce $1 Million Hackathon last month and we placed 5th! We won $15K and got to run around Dreamforce parties for free. Thank you Benioff.

Our mobile prototype visualizes relationships between any pair of variables in a Salesforce database. We came up with a way of calculating how interesting any data relationship will be, using the chi-squared test to compute the relatedness of two variables, like sales over time. We then sort all possible relations in the Salesforce database with this metric!

When you tap on one, we figure out the most suitable visualization and let you interact with it. You can favorite Salesforce objects and share visualizations with coworkers. There's zero configuration and instant insight discovery.

DC coded the math and made incredible d3 animations that adapt to arbitrary data types. I designed it and worked on the backend in heroku, integrating Force.com.

We'll be open sourcing it and releasing it onto the Salesforce AppExchange hopefully soon. Contact me in the upper-left if interested.

Selling software to hedge funds is a Catch-22: the technology is either so good that the best thing to do is start a new hedge fund, or the technology is just bad enough that funds won't feel uncompetitive if they share it with others.

The hedge fund world is a zero-sum game. Over the past summer, I learned just what this means for Silicon Valley. Many of the financial tech tools, web scrapers, and data analytics tools out there are viewed by a hedge fund manager like this: "If this tech was any good, it would just be a hedge fund. It's not a hedge fund, so it must be shit."

That's almost word-for-word what the founder of a large hedge fund told me over the phone.

So after just 4 months, my cofounders and I decided to put Apollo.ai on the shelves. Our technology helped predict a market opportunity that netted one fund $2 million. We had built a way to do automatic data integration over arbitrary data sources, using natural language processing to extract entities and create a universal ontology for a hedge fund's data. And we had started a partnership with a large data provider to give us a unique edge.

But the gold mine for hedge funds is predictive analytics, and specifically, the construction of the predictive financial models that they build themselves. Not exactly in the purview of a tech company like ours, which sought to avoid financial modeling so we could focus on design and data analytics. I guess I realized that I'm not a finance guy.

Special thanks to the great cofounders I worked with: Chris Guthrie, Armaan Ali, and Griffin Price. And another thanks to all of our advisors and friends we met along the way, especially our Lightspeed Summer Program advisors like Jeremy Liew, and Pejman Nozad and Mar Hershenson of Pejman Mar Ventures for helping us find housing and space. I feel bad we couldn't make it worth your time, but at least I learned more than I ever expected.

I put together an opinion piece for the Stanford Review on why technologists should experiment with hardware. As Apple and its alumni's companies continue to demonstrate, custom hardware provides a unique advantage for any product, allowing engineers to build a tailored portal to their software.

This past weekend, my friends Doug Safreno and Clay Schubiner won the “best API integration” award with me at the 2014 Meteor Hackathon. We built Saurus, a word processor with real-time natural language processing. It finds synonyms and Wikipedia articles for you as you type:

Saurus uses context-aware parsing and part-of-speech tagging to recognize named entities in order to provide smart Wikipedia results, word synonyms from WordNet, background themes from Flickr using AlchemyAPI, and more.

Check it out at http://thesaurus.meteor.com

I started working on a new Meteor project and wanted to catch more JavaScript errors at compile time, so I switched to TypeScript.

Unfortunately, Meteor recompiles every file whenever you make a change, which slows down dev cycles at scale and defeat the purpose of live reloads in Meteor. So I decided to help Olivier Refalo with the Meteor package for compiling TypeScript files, and the result is an optimized compiler that caches JavaScript and runs much faster, depending on how many files you have. (With around 10 files, after changing 1, you get a little under a 10x speedup.)

https://github.com/meteor-typescript/meteor-typescript-compiler

This is my senior project: a Chrome app that continuously scans for time-sensitive emails and sends you reminders automatically, so you don’t have to worry about them. Kind of like an AI-based secretary. It made it to finals for the BASES Product Showcase.

Key challenges here were in building a custom cross-platform notification center, using natural language processing (NLP) to parse dates, times, and timezones from cryptic clues like "1pm on mon," and using IMAP to push notifications from the server, securely, and in real time. In the first few hours, the Amazon server I had crashed under the visitor load from Facebook, so I spent a lot of time learning how to scale the product.

Here’s a link to the public beta on the Chrome app store; let me know what you think. All your data is now encrypted, the detected time displays on the notification, there's timezone and blacklisting support, and more.

...by forcing them to quit the browser and reload the site.

When you try to navigate away from any webpage, the "beforeunload" event gets fired. Most browsers have many security checks on it. Safari has fewer.

Try entering the console in Safari (Develop -> Show Error Console) and pasting in the following code (warning: leaving this page will then cause havoc until you force quit Safari):

window.onbeforeunload = function(e) { while(true) {alert('bye');} }

The code above creates a continues sequence of JavaScript alerts that can't be stopped until you force the browser to quit.

This unstoppable alert sequence has been known on the web for a while, so why is this any different? If you attempt to leave, force the browser to crash, and then reopen Safari, the browser will restore your tabs and you will be doomed to crash on that website forever. Because Safari loads normally each time after crashing, it doesn't think anything is wrong and thus restores your malicious tab.

Chrome avoids this problem by preventing beforeunload handlers from creating JavaScript alerts. Instead, the return value is used inside of a special Chrome dialogue that asks if you want to leave the site. In general, Chrome allows users to prevent a page from creating new alerts.

Safari doesn't even have post-launch tab preferences, so you can't prevent it from reopening the tab unless you can close it quickly enough.

I made a dynamic image generator yesterday to solve the annoyance of emails that get reply-alls with just "+1" or "yes."

It generates a clickable image with a "like" count that you can paste into your emails, so people can like them. The key trick here was adapting to the restriction that it can't use JavaScript or any kind of code: just an image and a link. That's all that's allowed in emails anyway.

It also tracks how many people view your email, among other statistics, with options to customize the button. The limitation is that the receiver may get prompted to allow images from you if you aren't in their contacts.

Campaign Trail is an online organizer I'm working on for social movements and campaigns. Rohan Ram and I are developing the product for use in Don Bivens' 2012 U.S. Senate Campaign in Arizona.

Here's a screenshot of the newsfeed I just finished:

Users can create projects and events with assignable tasks, ask and answer questions about the campaign, contribute pictures, articles, and other media, interact with moderators, message other users, and more.

What follows is a design overview and a technical overview, with more screenshots at the end.

Dormlink is a social network that I made with with Bryant Tan. It's devoted to college students in dorms, currently in operation at Stanford for 5 freshman dorms and 180 freshmen. The Stanford Daily wrote about it here.

We made the front page!

I designed the site, built a php framework, and constructed the database. It's now being hosted on a server in my dorm room. You can try it out using the demo link at the bottom of http://dormlink.me/.

In the summer of 2010, I interned at the Trade Partnership in D.C. to make a map application for showing U.S. businesses that benefit from a trade program. The firm (tradepartnership.com) wanted Congress to renew the program, so I helped lobby the application on Capitol Hill. The experience culminated in a cool trade dinner with senators whose speeches put everyone to sleep.