|
|
B.A. (Hons) First Class, M.Phil. and Ph.D. in Computer Science/Natural Language Processing from the University of Cambridge, England. Life-long keen interest in mathematics, computing and science.
Versatile, with a background in IT development in the internet, banking and telecommunications sectors, and an academic background in computational linguistics.
15 years of professional employment experience, including 8 years of C++ and 2 years of Java programming on Unix/Linux.
Main recent skills and technologies: Java, JBoss, C++, OO Design, STL, Boost, Multithreading, SQL (Oracle, Sybase). Experience with C#, HTML, CSS, IBM WebSphere (MQ) messaging, Perl. Extensive Lisp experience from the Ph.D. years.
Fluent in English and Norwegian, intermediate level in Catalan and German.
Migoa is the company behind the property market-oriented vertical search engine Nuroa (nuroa.es, nuroa.de). My main role is in the research and development for the crawler and web server back end, but as this is a relatively small internet start-up company, I am also regularly involved in the front end development and other all-round IT tasks.
The systems are written in Java, running in a JBoss environment, and with heavy dependence on Oracle databases. The technical challenges involve the retrieval and analysis of property market ads from hundreds of web sites, the efficient storage and management of the data extracted from these pages, a search function based on a combination of semantic values and keywords, and the user-oriented web page interface to this search engine.
Morgan Stanley is a highly regarded global financial services firm, with over 54 000 employees in over 30 countries, and with large IT departments in New York, London and Tokyo. It was ranked number 20 on the Fortune 500 list of the largest corporations in the USA in 2007.
I was working in a team of 8-12 people divided between London, New York and Tokyo, on a project developing a new trading system for fixed income repo trades. This was a large development and was gradually being phased in as a replacement for a legacy system. The system supported a wide range of operations required for this business, but its core functionalities were to keep trading positions (based on trades and other data received in high-volume feeds from a central mainframe system), to allow users to enter new trades, and to provide semi-automatic pledging of collateral for the fixed income tri-party business. The system was of critical importance for the bank's business, and was therefore developed to the highest standard of reliability. The pledging system, for example, was used for covering trades to a daily total of $220 billion in New York.
My involvement was mainly on the back end, which was implemented in C++ and Sybase and consisted of several multithreaded server processes communicating with each other, the mainframe and the front end with XML over WebSphere MQ, SOAP and a proprietary pub/sub framework. It was initially developed for Solaris, but was later primarily used on Linux servers. There was also a gradual move from in-house libraries to extensive use of STL and the Boost libraries in the course of the development.
In the last year of this job I was taking on a leading role, as the most senior developer on the project in the London office. Besides my technical roles, I was also involved in recruitment, including the selection and interviewing of job candidates.
I played a key role in the development of the above-mentioned semi-automatic pledging system, in particular on designing and implementing the logic for selecting securities for use as collateral based on customer and business requirements defined in XML. This logic sought to optimize the choice of securities in order to meet the customers' requirements while simultaneously preserving the firm's best interests.
I wrote a credit exposure reporting engine for generating reports of the firm's exposure from all the repos and tri-party allocations in the system at any time. These reports were needed to meet new regulatory requirements, and were therefore developed to a strict deadline. The main challenge was to calculate repo interest and exposures according to complex business rules reflecting local conventions in the three main markets (New York, London and Tokyo).
I also wrote a risk reporting engine for generating reports of interest rate-based risk metrics calculated from the repos and tri-party allocations. Again, these reports were needed for meeting new regulatory requirements. The challenges were partly similar to those of the exposure calculations, but included also a detailed analysis of the timed cashflows associated with each repo. My solution included an innovative dependency-driven algorithm, structuring the computation by the format of the database used for delivering the results.
At the end I was working on migrating the pledging system to London, extending it for the local business practices and requirements. Meanwhile I was supervising ongoing work on extending the credit exposure and risk reporting engines to cover additional business areas, and provide related data to other business units.
Logica (known as LogicaCMG since a 2002 merger) is an IT consultancy firm, specializing in a number of areas including telecommunications, systems integration and military applications. It has currently over 30 000 employees in 36 countries.
I was working in the telecommunications division of Logica, and was mainly involved with two projects concerned with a customer management system and a billing infrastructure for mobile telephone companies.
I designed and implemented an interface between a Customer Care and Billing System (FROST) and a Geographical Information System (Netgraph) used by the Polish telecommunications company Netia for managing the development and usage of their telephone line network. The interface was to enable FROST to check the availability of lines before new sales, reserve such lines at the time of a sale, and activate them when a sale was completed. The work required several visits to the client's offices in Warsaw. The development used C on UNIX and SQL.
I designed and implemented an ASN.1 Decoder module for a system for collecting Call Data Records from a Motorola GPRS network, acting as a Charging Gateway between this network and a Customer care and Billing system. This program was written in C++ and used Snacc for the basic ASN.1 decoding/encoding.
After completing my Ph.D., I stayed in the university as a post-doctoral Research Assistant to help prepare a new release of their comprehensive Natural Language Processing system, the ‘Alvey Natural Language Tools’ (ANLT), which would include the program I developed for the Ph.D.
Norsk Data was a highly successful computer manufacturer in Norway in the 1970s and 1980s, specializing in minicomputers for research, public administration and education. It collapsed in 1992 when it was unable to adapt to the general move away from proprietary systems to PCs and Unix at the time.
I worked on a project which, in collaboration with other companies and research organizations, developed a system for processing Synthetic Aperture Radar (SAR) image data from the ERS-1 satellite. The system was for a new ground station of a Norwegian unit associated with the European Space Agency (ESA).
Norsk Data was responsible for the software for the high volume data transfer, and for the system integration on Norsk Data mainframe hardware.
I participated in the planning, design and detailed design phases of the project (the implementation happened after my departure).
I took part in the technical coordination between the Norsk Data team, a specialist consultant (for the SAR computation) and a team at the University of Tromsø (for designing the ground station), travelling extensively between these participants.
I was responsible for many ESA-compliant reports required at specific stages of the work, including the system design document written to rigorous ESA documentation standards.
My Ph.D. thesis, Statistical Parse Selection using Semantic Cooccurrences, was about probabilistic disambiguation of naturally occurring sentences, based on cooccurrences of predicates in semantic forms and statistical data of such cooccurrences in a parsed corpus. I used the Wall Street Journal part of the Penn Treebank for training and test data, and developed a medium wide-coverage, computationally tractable, syntactic and semantic grammar to provide an accurate set of possible analyses of most sentences in the corpus. The work involved solving many problems inherent in corpus processing, such as building coherent lexicon files to match the data and the grammar, coping with irregularities in the data and unknown words, etc.
My solution was based on an original probabilistic model of the data and used Bayesian estimation (based on probability density functions carefully derived from scratch) for maximal precision in computing the fundamental probabilistic parameters of the system. The fact that these estimates are statistically unbiased eliminates the need for smoothing with what is highly sparse data, further improving the overall precision. The solution was implemented in Lisp.
This is a one-year taught M.Phil. degree covering a wide range of NLP-related topics: introductory linguistics and phonetics, major study of speech analysis (recognition and synthesis), syntactic parsing, semantic interpretation, text and discourse processing; some psycholinguistics, experimental psychology, speech and language applications. The course lasted twelve months, including a project with a 15,000-word dissertation: Extended LR-Parsing of NL Text: Applying Bermudez's LAR Parser to Natural Languages.
This is the standard Computer Science undergraduate degree in Cambridge. I obtained First Class results each year, and was also awarded a First Class for the final BA degree. (British universities give a final grade for the whole degree course; Cambridge, in addition, gives an overall grade for each year of study.) My examination results put me in the top 2% of Cambridge graduates.
This was originally awarded to fund my Ph.D. studies at the level of a standard British doctoral grant, but was reduced to a minor annual sum in view of my more substantial funding from Norway for the same period.
From the Committee of Vice-Chancellors and Principals of the Universities of the United Kingdom, it covers the difference in tuition fees payable by overseas students and home students. It was awarded to me in 1991 and renewed for the maximum two further years in 1992 and 1993.
This grant, in the form of a standard Norwegian researcher's salary, was the main funding for my Ph.D. studies. It was awarded to me in 1991 and renewed for the maximum two further years in 1992 and 1993.
This scholarship was awarded to me for two years on the basis of my examination results.
“Probabilistic Parse Selection based on Semantic Cooccurrences.” In Harry Bunt and Anton Nijholt eds. Advances in Probabilistic and Other Parsing Technologies. Dordrecht, Boston and London: Kluwer Academic Publishers, 2000. Originally presented in The International Workshop on Parsing Technologies, MIT, Massachusetts, USA, in Sep 1997.
| Norwegian | Native, able to speak or correspond with speakers of Swedish and Danish |
| English | Near-native (resident in the UK for 19 years) |
| Catalan | Currently studying in last part of intermediate-level course |
| German | Intermediate |
Driving Licence: Full clean British driving licence.