Congratulations to all the teams for making it to the finish line, overcoming the 36 hour pressure cooker and coming out with a lot of fabulous ideas and compelling demos. I was also put under extraordinary pressure to make a decision for the Open Data workshop award - and I attribute the fact that I was using data to track and quickly calculate essential variables to keeping my senses in the final minutes. Congratulations again to team InvestoBot!
Many of the 14 teams I interviewed (out of 40 overall) were, in fact, exemplary in their use of open data and thoughts about publishing data and creating platforms; it goes without saying that, if I could, I would have given all of you a prize. But I trust that what you have learned, connections made and the experience gained will, in the long run, be the best reward.
In follow-up to the hackathon I had another interview with the InstaBot team. Here is what they had to say about the experience:
InstaBot's team first got together when theme of hackathon was announced. Each of the members of the team is quite interested in the stock market - something that thousands of people participate in, without the possibility of completely understanding the rules or the possibility of predicting outcomes. They say the finance theme is special, few hands-on events like this exist - usually hackathons are open themed, and it was their first hackathon working as a team.
Why do you care about Fintech? Power shifting to the community rather than centered on large corps. Smaller players controlling banking landscape in the future. They work in the belief that "decentralised banking" can lead to more innovation. One of them already made a video about Artificial Intelligence in financial industry during a course. They are Computer Science students interested in machine learning, something that needs a lot of data.
They learned a new library for their project, had an excellent experience, and the teamwork worked great. But unfortunately there is no time for them to make a startup now, as they are first year students, only in London since September.
The project's backend code is in Python and open source. They get some runtime data from IBM, otherwise the code is self running. They took a look at some neural network papers, accuracy 30%. Through Google Scholar, found some stock market prediction. Started using sentiment analysis with Bluemix/Watson, and Google Finance for stock prices are directly news connected (top 10 articles about the company stock).
The founding premise of their project was developed based on the idea that the data is open. Small organizations are relying on this kind of data, and they imagine that making it private would be a disaster that wouldn't just affect their hack. They also looked at Yahoo Finance and mentioned that a lot of other APIs can a similar job. Many NASDAQ companies even publish their own data using their own APIs.
They were inspired by the journalist from WIRED who spoke at the start of the hackathon. He talked about how this could change a lot in user interaction - a powerful idea...
They didn't have time to look at standards, were too focused during the hackathon. Some of the other challenges: they started with Facebook Messenger (private chat), the API was not working, after 4-5 hours gave up. Twitter's API worked much better.
They had accuracy problems: really high accuracy was giving from the sentiment API but not possible, they didn't take the sentiment initially into regard. Confidence was too high because the network started overfitting the data, and time based prediction was "too perfect". This was important because in all the papers they looked at the rate of success is based on the back-predicting. They said that getting started with Bluemix/Azure is quick but the UI is messy. It takes a long time to find what you need & set up the service. Took us much longer to find on the website where to find the username and password (even the helper from IBM had trouble finding it).
But.. on Saturday afternoon everything was ready to test, they finished work before midnight and started preparing for the pitch. After debugging so many issues it was time for the 2 minute pitches.
What made your project a success? "We tried not to aim it at similar people to ourselves (i.e. same background). Our user is anybody who doesn't have a lot of technical knowledge, just a normal person who is on the go and wants to invest." They want quick short info on their phone, and this idea resonated with some people. If people can see that our prediction is accurate, we would influence their decision.
Next steps: a subscription model, we would give you extra data, private message capabilities, empowering the investor rather than invest on their behalf. The team is based in London and surely can develop it further, have access to the resources, learn more about the market, how our startup could function, what would be the first steps in developing a company around the bot. The IBM mentor was constantly asking technical questions, there was initial encouragement but besides tips on how to improve it, we didn't get much business advice.
"We'll see, after the exams..."
Good luck, guys!