Genady Yoffe Email & Phone Number

Chicago, Illinois, United States

Algorithmic Trading Strategy for confidential client:
• Conceptualized, implemented, deployed, and maintained short-term sentiment analysis-based trading strategy.
• Combined NLP-analyzed social media sentiment with stock performance data in stats models to enhance trading decisions.
• Strategy generated over $1.9M in profits during peak periods and achieved Sharpe ratio of 2.4.
*************************************************************************

Logistics Optimization for Robertson's Ready Mix:
• Crafted sophisticated multi-step constrained optimization algorithm for ready mix delivery scheduling.
• Led team of data scientists to implement, deploy, and maintain pilot solution utilizing 120+ trucks (based on designed algorithm).

***************************************************************************

Image Document Analysis for InterFirst Mortgage:
• Engineered deep learning ensemble for mortgage image document classification.
• Achieved 85% accuracy across 50+ document classes by merging visual and textual analyses.
• In addition built and deployed complex ML pipeline for extraction of financial data from diverse mortgage documents submitted as images.
• Addressed challenge of non-standardized layouts and terms by utilizing NLP and advanced statistical models.
• Extensively employed computer vision algorithms to improve image quality prior to OCR application.
• Cumulative impact of automation of document classification and data retrieval reduced manual document processing by 50%, effectively halving loan application validation time from two days to one.
***************************************************************************

Predictive Maintenance for Robertson's Ready-Mix:
• Developed and deployed predictive models for internal heat of motors in rock plant equipment.
• Achieved 90% prediction accuracy; models' rollout prevented equipment failures and ensured continuous operation of 7 rock plants.

Greater Chicago Area

• Improved ‘loanee payday’ model – to effectively set loan payback days (accuracy increase from 90% to 92%).
• Created ad-hoc payroll recognition model – to identify payroll stream in applicant bank statements.
• Established several fraud filters (which helped to decrease cost per funded by 15%).
• Created innovative features and optimized feature selection for customer eligibility model.
• Eligibility model enhancements secured more than extra $500,000 in loan repayments.
• Built AWS-based data pipeline.

• Maintained company's signature product for forecasting demand and establishing price elasticity estimates across high-profile CPG/Retail clients.
• Led consultant teams to adapt, tune, and customize products to align with clients' unique requirements and data specifics.
• Tuned solution upped 5-10% goods allocation throughout stores for multi-billion retailers.
• Designed, implemented, and productionalized new algorithms to enhance company's solution for forecasting demand.
• That includes 1) demand cannibalization impact functionality, and 2) incorporating advanced shrinking/selection multiple regression methods for establishing relevant demand drivers.
• These algorithmic enhancements improved overall solution’s prediction accuracy by 1.5-2%.

Chicago

1. Developed a C++ application for reconstructing complete pricing history of stock securities. The program cross-references multiple data files from different vendors (such as Bloomberg corporate actions feed, SunGard pricing files, etc). It analyzes corporate actions in order to accurately identify stocks of the same financial entity throughout its entire history, regardless of merger/acquisition-related changes in the identifiers. The program calculates split-adjusted historical prices and feeds them into company’s pricing database. The deployment of this program significantly extended company’s pool of securities and improved accuracy and scope of its quantitative research.

2. Redesigned company’s pricing database by merging multiple separate databases onto one comprehensive central resource. The new design eliminated pricing inconsistencies across the firm, saved space, lowered cost and reduced maintenance overhead while maintaining same or better query performance and output.

3. Developed a robust C++ application for creating internal firm-wide security identifiers. The program utilizes Number Theory in order to quickly assign to securities unique names of close to shortest possible length. The generated identifiers easily map to the standard identifiers used by all major exchanges, yet they are significantly shorter, which allows for faster and more convenient quantitative research operations.

Chicago

Designed and developed massively parallel implementation
of QR factorization via Gramm-Schmidt orthogonalization for matrices of lower dimensions (about 32) on NVIDIA TESLA 2050 using CUDA framework.

Chicago

Data Mining: Cluster Analysis

 Created an efficient model for extracting elevation profiles from noisy and biased data for a new GPS product. The model was accepted by the company for the future development as a prototype of the product.

 Implementation: used R and igraph package.

 Used in-depth mathematics, statistics and computer science algorithms.

 Further details of the prototype cannot be disclosed due to a signed NDA agreement.

 See a recommendation by Navteq’s executive on my LinkedIn profile.

Applied Numerical Analysis at MCS
Development of multigrid methods for solving nonlinear partial differential equations (C/C++)
(see technical report on the personal website for the obtained results)