Just for the record, I left MyRoar (where I was CTO) last week.

Applying NLP to financial information was very interesting and I really enjoyed the challenges that presented. In the end there were too many creative differences between Kate McDonough (the original founder) and myself, and I felt I could no longer be an effective CTO to the company.

MyRoar was written up in Mass High Tech today:

What if you could ask the web a question — and get an honest answer?

If you’re a financial analyst, a Boston startup called MyRoar Inc. says you can. The bootstrapped venture is building a semantic search product designed to answer analysts’ questions directly, like, “Is anyone backing this bond?” or “How many hedge funds are there in the U.S.?”

A couple of weeks ago I decided to upgrade my main linux development machine, a four year old dual-Xeon machine, with something a little more modern and much less power hungry. Indeed the old machine accounted for between 25% and 30% of my power bill at home.

I did not need to upgrade the entire machine, just the motherboard and the memory. I opted for an Intel DX58SO board with a Core i7 processor. In the event I did need to get a new DVD drive, the DX58SO does not have any legacy ports (so no PATA connector), and a new graphics card, neither of which were all that expensive.

I did run into a couple of snags along the way. I use CentOS 5.2 as my default Linux distribution (I developed a taste for boring and stable after a brief detour into cutting edge, exciting and unstable with Gentoo a while back.) The bummer was that the onboard network adaptor is not supported by CentOS 5.2 and I would need to download and install the appropriate driver, which struck me as a step in the wrong direction, so I opted for Fedora Core 10 instead which I am very comfortable with.

Somewhere along the process the power supply decided to commit Seppuku with a loud bang, a big blue flash and a small cloud of acrid smoke so I had to run out and replace that too. I was glad that it did not take out the motherboard in the process.

So now I have a new machine with 6GB of RAM, 4 cores which look like 8 cpus to Linux (Hyper-threading trickery), which is a lot quieter, a lot more power efficient and a lot cooler (summer is approaching) than the previous noisy space-heater I had. The quietness alone is worth the price of admission.

MyRoar was written up twice recently. On the New Ideas Engineering weblog:

This year we also want to mention a newcomer in search, MyRoar. Formed by technical folks who understand the financial services market, MyRoar is a natural language question machine. They claim that their technology does not need to be trained like vendors InQuira and Q-Go; we’re hopeful to do more with them this year so we can report back on our findings.

And on Stephen Arnold’s weblog:

A happy quack to the reader who alerted me to MyRoar.com. This is a vertical search service that relies on natural language processing. I did some sleuthing and learned that François Schiettecatte joined the company earlier this year. Mr. Schiettecatte has a distinguished track record in search, natural language processing, and content processing. French by birth, he went to university in the UK and has lived and worked in the US for many years.
.
.
.
You can use the system here. The system performed quite well on my test queries; for example, “What are the current financials for Parker Hannifin?” returned two results with the data I wanted. I will try to get Mr. Schiettecatte to participate in the Search Wizards Speak interview series. Give the system a whirl.

About 2 months ago I took the position of CTO at MyRoar, an NLP based financial search engine founded by Kate McDonough.

I joined because it was a good fit and there was a lot of overlap with the project I was working on, specifically in the back end needed to drive a search engine. They had the NLP engine and I have everything else around it.

After a slow start I effectively took over development over Thanksgiving weekend, getting a demo site running in about four days and am currently working on refining the demo, getting more data into the index, and working with Kate on raising the funds needed to take this to the next level of development.

The demo is pretty small, on purpose. It is a demo and it was put together in very short order. Since then I have been refining it to grow a little more and to index more data. For me is has been a very interesting learning experience about NLP, and the performance characteristics of NLP which are, shall we say, interesting.

Along the way I also took over the development of the front-end, we also have a mobile front-end. I had to learn enough PHP to muddle my way through in a few days, and it is not perfect but it works well enough as a demo.

Currently I am looking into using Amazon’s EC2 for processing, because NLP always needs more.