Cold Vault - The motivation

We waste a lot of energy in our homes. One of the biggest energy hogs in a home is the refrigerator. The efficiency of fridges have improved dramatically over the last twenty years, but they are still frequently the largest consumers of electricity in houses.  The fridge needs a radical rethinking to make it sip energy instead of gulp it.

Translation Hierarchy - The pitch

Why settle for extremely expensive translations or inaccurate machine translations? With Translation Hierarchy, you get the best of both worlds: affordable translations at near expert accuracy.

Translation Hierarchy - The fun part

I like what can be done with Mechanical Turk-like systems. There is also a fun optimization problem here: how do you best choose the amount of work done at each level. Also, expert translations can be given as feedback to intermediate translators, which could improve their skills.

Translation Hierarchy - The growth potential

I think that there are hundreds of thousands of domains that would want to have proper translations. I think the cost of a translation would be around $500 on average and that the revenue could turn out to be $10 million a year.

Translation Hierarchy - The monetization

Let’s take an example of the domain of a large city, say San Francisco. Suppose there are eight thousand pages within the domain. If the machine translation confidently translates 95% of the pages, then 400 pages go to native speakers at a cost of 25 cents per page. If half of those need to be re-translated, then 200 pages get translated by humans at a cost of one dollar per page. Of those pages, say 40 translations have disagreement and go to an expert translator. If the cost of an expert translation is $5 per page, then the total cost is $500 for a high quality translation of the entire domain. The domain could also specify which pages on their domain need to be of higher quality than others. Some very important documents may be critical to have translated perfectly where other general information pages may be fine with a few small errors.

Translation Hierarchy - The idea

The most accurate way to get translations is to hire someone who is fluent in both the source and target languages to do the translating. Unfortunately, this is very expensive and doesn’t scale well; After all, there aren’t that many people fluent in many languages. The other extreme is machine translation, like what Google does with Google Translate. This option is extremely cheap (free actually), but it’s not very high quality.  What is really needed is something in between.

The best way solve this problem is use a hierarchy. All pages on a domain would get translated by machine translation. Each machine translated page would come with an estimate of how accurate the page is. One way to do that is to train several language models on subsets of the training data. Each model does a translation and the confidence of the translation is the amount by which they all agree.

Pages that had low confidence translations would then be sent to native speakers of the target language.  It’s usually fairly easy to see if a page is poorly translated without knowing the source text.  One benefit of using target language speakers is that they don’t need special training to perform their task and can be fairly cheap. If the native speaker finds any problems, they would then escalate the translation to humans (intermediate translators).

The escalated translation is handed to two translators with some (but non-expert level) experience. If both translators independently come up with the same translation, that text is used. If they disagree then it’s escalated again to expert level translation. Only a single expert is needed to translate a page. By using experts only in the rare cases when they are needed, costs can be kept low while quality is kept high.

Translation Hierarchy - The motivation

Getting good translations is hard. Google tries to do it, but it rarely makes great translations.  As more non-English speakers come onto the web, proper translations of web text become even more important.

Fresh Test Sensors - The pitch

Are you sure that milk is as fresh as it says? You can’t really know unless it comes with a Fresh Test Sensor. Fresh Test Sensors are ever vigilant guards against groceries getting exposed to the elements.

Fresh Test Sensors - The fun part

This will require some real ingenuity. Either the chemical or electric route would require being very clever about making something that works and is also extremely cheap. This idea is only viable if the cost at scale can be much less than one cent per sensor.

Fresh Test Sensors - The growth potential

Let’s start with milk. Let’s say there are 100 million households in the US and on average each one buys a gallon of milk a week. If each had one of these sensors, that would be $1 million a week or $52 million a year in revenue. If this grew to a number of other grocery and pharmaceutical goods, it could grow to be ten times that.

Fresh Test Sensors - The monetization

The sensors would need to be extremely cheap to manufacture, roughly one-tenth of a cent. The sensors would be sold for one cent per sensor.  For a gallon of milk, adding 1 cent doesn’t have a real cost impact, but could give real piece of mind to consumers.

Fresh Test Sensors - The idea

It should be possible to create extremely cheap sensors that tell if a product went above a given temperature. For example, there could be a chemical sticker that changes state (and color) when it goes above a given temperature, something similar to thermochromism, except the change would need to be irreversible.

Chemical Sensors

This would require developing a cheap non-toxic material that makes an irreversible state change when it goes above some temperature T. The chemical would be put into a sticker on goods like milk cartons, egg containers, and other refrigerated products.  The simplicity of a chemical change also, however, comes with a lack of detail. The change only tells you that it exceeded a temperature, not how long the temperature was exceeded, which may be an important factor.

Electrical Sensors

The alternative is to make cheap chips that have temperature sensors and a tiny battery. The chip would spend 99% of it’s life sleeping conserving battery live. For a second every hour, it would turn itself back on and measure the temperature. The hourly temperature would be recorded until the good arrives at the final point of sale (the grocery store). There a grocery worker would use and RFID scanner to scan the package and see a summary of the temperature history of the good. If it doesn’t fit the federal, state, and store policy, then the good is rejected. Also worth noting is that other sensors could also be used to measure exposure to sunlight and acceleration.

If implemented at scale, these sensors could ensure food safety is cheap and simple. This could also apply to imported and exported goods.

Fresh Test Sensors - The motivation

One challenge with grocery goods and pharmaceuticals is that they can become spoiled if they spend a certain amount of time above a given temperature. From the source to final sale, it can be hard to determine if the goods stayed cool the entire time, as there can be many handoffs. It would be nice to have a simple, easy to read indicator that a product is still good.

Empowering the Deaf - The pitch

Life is hard for the severely hearing impaired. They cannot easily communicate with people around them. With the Empowering the Deaf system, the barrier simply vanishes. The signing is turned into speech projected by the smart phone and speech from the other person is converted into text viewed on Google glass. The result a truly seamless communication experience.

Empowering the Deaf - The fun part

This is a great project for combining software and hardware skills. Getting everything integrated into a product that the user doesn’t even notice is the end goal. There are many ways to increase the usefulness of the product, including better speech recognition and sign recognitionas well as better synthetic voices, to name a few.

Empowering the Deaf - The growth potential

There are roughly one million Americans who are functionally deaf.  If half of them bought a unit, and upgraded every 5 years that would be 100,000 sales per year or $50 million in profit.

Empowering the Deaf - The monetization

The hardware alone for this product will run in the $3,000+ range. I expect that a total package would sell for $5,000, with a $500 profit per sale.

Empowering the Deaf - The idea

The goal of Empowering the Deaf is to leverage technology to break down the barriers between the deaf and non-deaf as well as improve safety through situational awareness.

Communicating with the severely hearing impaired usually requires both parties to use sign language. Usually the hearing impaired person will know how to sign, but that skill is rare in the general population. The solution is to have each party communicate in the way that is most natural for them. This requires real-time speech recognition technology and a way for the hearing impaired person to see the text, something Google Glass would be great for.  This wouldn’t necessarily be as hard as it sounds. What is required is a lapel mic that can pick up the other person’s speech, a smartphone with the speech recognition software, and a bluetooth connection between the smartphone and glass to show the text.

Now, let’s cover how the hearing impaired person will communicate with the other person. Again, another budding technology will play a key role, this time it’s Leap Motion. Next to the lapel mic on the hearing impaired person’s chest will be a Leap Motion sensor. This sensor is designed to accurately measure hand and finger movements in 3 dimensions, which is perfect for translating sign language into speech. The user’s smartphone will again play the intermediary between the sensor and the output, which this time would be a speaker that will say what the person is signing.

I think the above is enough to have a ground breaking product, but I think it could do more. Everyday we use hearing for more than just listening to people, we also use it to navigate our environment. If the hearing impaired user had not just one mic on them, but several, they could be used to signal to the user what sounds are around them and the direction they’re coming from. It could be very valuable to know that there is a car sound to the left and another car sound behind the user. Knowing what’s going on around the user could not only increase their safety but also their general quality of life.

Empowering the Deaf - The motivation

Losing a sense can be debilitating. For example, being deaf make communicating with most hearing people around you very difficult. It can also make navigating urban environments extra dangerous, i.e., not hearing aproaching cars and other motor vehicles.

Virtual Scout - The pitch

Film study can be drudgery. Hours and hours spent just figuring out which plays the other team does on average. Then comes drilling down in individual player performances. There is a lot of manual work that doesn’t need to be done by people. Virtual Scout make film study easy. Want to see all examples of 5 yard pass plays? Easy. What to know what percent of the time on 2nd and short they throw deep? Simple. Virtual Scout lets coaches focus on the real brain work in film study.

Virtual Scout - The fun part

I love computer vision, and this project has it in spades. Not only is 3D reconstruction an interesting problem, really good reconstruction could help determine if players have a tell, like in this story.  Then, there is also the statistics aspect. For example, knowing that when previously in a particular position the other team used one play 80% of the time is really powerful.

Virtual Scout - The growth potential

There are 32 NFL teams and over 100 FBS college football teams. If 100 teams got licences, that would be $2.5 million a year in revenue. Smaller football teams and even some high school teams may want licences. For a stripped down feature set, the product could be offered at $5,000 a year. Even with 1/5th the price, there could be five or ten times as many teams getting licences. This would grow the revenue to $5 million a year.

This could even expand beyond football to... well football (or soccer as we call it here). There are hundreds of football clubs large enough to want to use this software for their own film study. I expect that expanding to other sports could grow the revenue by another factor of 10 or to $50 million a year.

Virtual Scout - The monetization

The product sold to teams would be the software that makes the film database and analyzes each play and adds metadata about it (such as the formations and the outcome the individual player behavior). Each team would get a licence for the entire organization, which would give them access to cloud software.  They would upload the film, where it would be processed and available to everyone on the team to analyze. Interestingly, this software has a natural piracy deterrent: giving someone else a password gives them insight into what film you’ve been looking at and any notes you’ve added to different clips. Both college and professional football teams spend a lot of money on facilities and practice equipment. Compared to these other costs, $25,000 a year for a licence isn’t very much.

Virtual Scout - The idea

For college and professional football, there are a lot of video cameras at each game. The number of cameras vary from 5 to 62 per game. I couldn’t find a great source on this, but it looks like coaches get access to at least 2 film feeds (if not all of them). The two that the coaches get have views of all of the players on the field. Thankfully, two views of the players is all that is needed to make a 3D reconstruction of the playing field.

With the location of every player at every second of the game captured. It is then very easy to characterize each formation and play that happened. Suddenly, no work is required to get the distribution of formations a team uses or plays they run. If several weeks (or a year) of data is processed it’s possible to estimate the probability they will run a particular play in a given position on the field.

This can even give insight into individual players’ performance. How often does this defensive tackle get penetration and cause the running back to alter his path? What is the average amount of separation this wide reciever gets? Of the missed passes, how many are on the wide receiver and how many on the quarterback?

In addition to processing the film for statistics, this software could make a database of different plays a team has done. For example, a coach may want to show his players how much success the other team has had with the jet sweep. Also, the coach may want to see the common plays the opponent uses on 3rd and long where they actually get the first down. Practically all aspects of film study can be improved with automation.

Virtual Scout - The motivation

Football coaches spend a lot of time looking over film of opponents’ previous games. From the film they hope to understand which plays the other team likes to use in which situations. They also hope to get a sense for how the players react under different scenarios. All of this can be an extremely tedious process. This job could really benefit from some automation.

Lean Green Gems - The pitch

There are three ethical downsides to buying jewelry: mining is one of the worst things you can do to the environment, many gems are involved with conflict, and jewelry (even high-end) is usually made in poor working conditions in developing countries. Wouldn't you rather wear something that aligns with your values? Lean Green Gems jewelry is beautiful, sustainable, made in America, and completely custom-made for you.

Lean Green Gems - The fun part

The metallurgy and chemistry makes this project fun. What sort of materials make for the best stones? Prettiest stones? Most durable stones? Also this project would include advantages of lean manufacturing to reduce costs, which is also cool.