- Principal Data Scientist2015 - PRESENTUangTeman
Technologies: Python, Amazon AWS, Linux
- Created and maintaining various statistical models for a lending company—the aim was to automate all aspects of the company's lending operations.
- Designed and implemented several successful credit risk evaluation models. Notably, I built models that supported big data—they made inferences on heterogeneous records.
- Designed and implemented statistical identity verification tools (big data style).
- Built scrapers for various social networks.
- Developed an OCR-based identity verification module.
- Made various further prediction models for collection, residence verification, and more.
- Research Contractor2015 - 2017Furukawa Electric
Technologies: Audio Analysis, Fast Fourier Transform Algorithm (FFT), Filters, Python
- Built a pilot system for direction detection using sound; specifically in an outdoor setting. The system's intended use is to enhance existing radar technology in the automotive field. Accuracy is lower than that of radar, but audio is a low-cost means to an orthogonal measurement.
- Developed a solution was able to measure the direction of passing cars, as well as show a consistent direction for the car it was mounted on.
- Ideated also physical solutions for alleviating wind distortion.
- Researcher2015 - 2017General Electric
Technologies: Audio Analysis, Fast Fourier Transform Algorithm (FFT), Filters, Python (NumPy)
- Delivered a high-accuracy direction-detection system using sound only.
- Fused the sound and video sensor data for enhanced motion detection.
- Built a successful traffic counting and classification system that was able to count passing vehicles and separate buses, cars, and more.
- Used Python (NumPy) to build the systems.
- Algorithm Developer2015 - 2015Analogy.co
Technologies: NLP, Python
- Built natural language models (NLP) for semantic data analysis.
- Developed automated semantic tagger modules for deriving meaning in the text.
- Used various algorithm optimization techniques for asymptotic speedup of the semantic search.
- Algorithm Developer and Programmer2011 - 2015Applied Logic Laboratory
Technologies: NLP, C, C++, Python, Java
- Built a syntax parser specifically for searching through English-language patents.
- Developed an information-rich semantic representation over the parsed syntax.
- Created a search engine for matching semantic information.
- Built scrapers and built scraped data into a structured dictionary automatically.
- Research Intern2006 - 2006SAP
Technologies: C, C++
- Created a server-side log analyzer. The log analyzer's purpose was to predict the identity based on user input.
- Designed the log analyzer so that it parsed server logs into structured data and estimated their distance—delivering probabilistic results.
- Algorithm Developer and Programmer2004 - 2005Bioscrypt Corporation
Technologies: C, C++, Embedded C
- Participated in building the company's fingerprint recognition algorithm. The algorithmic work included image cleansing, filtering, and feature extraction.
- Wrote parts of the matching algorithm.
- Developed exported APIs to the system.
- Created test tools.
- Programmer | Inventor2000 - 2002Connexus Corporation
Technologies: Audio Analysis, C, C++
- Invented one of the world's first content-based music recognition systems, it came earlier than the current market leader. The system monitored a large number of radio stations (US), to deliver high-accuracy recognition information of the contents, songs, commercials, recorded interviews etc.
- Designed the system so that it made heavy use of the Fourier transform to convert audio data into a musical score.
- Wrote fast algorithms that were supplied to match a time sequence of these fingerprints.
- Built automated management tools for recording and tagging unknown patterns; inserting them into the central database and various other smaller components.
- Research Intern1998 - 1998IBM
Technologies: Audio Analysis (FFT), C, C++
- Built a pilot program for converting an audio signal into musical notation. The criterion was for the musical score thus derived can be used for regenerating the original signal in a way that humans would still recognize it.