Nicolas Nicolov Email & Phone Number

Durham, Nc, Us

Focusing on identifying the right problems to solve (strategy) and creating the path to the solution. Building and growing advanced technology teams developing end-to-end tax compliance solutions with novel AI-ML components providing measurable and lasting impact to Avalara and our clients.

AskVera is your Virtual Environmental Recycling Assistant.We are working on initiatives at the crossroads of sustainability, AI/ML, image classification, object detection, text extraction, conversational dialog systems.

Boulder, Co, Us

Apollo is a compensation app, targeted to disrupt the $1.2 billion salary survey market. The focus is on the tech industry initially, and the system leverages Machine Learning engine and optimizes the financial efficiency of job offers.+ Setting vision, roadmap and execution. + Team building.+ System design.+ Predictive modeling: base/bonus/equity models (private and public companies).+ Overall quality.+ Model accuracy.+ Performance testing.+ Product, planning.+ Project management.

Organize the world's recipes, enable users to cook them in a heterogeneous connected environment (IoT), and monetize through grocery integration. Drop is building the operating system for the kitchen — one unified solution that connects the whole cooking journey.• Managing Data Science and Content teams (consisting of data scientists, engineers, data science content manager, content manager, content editors, technical advisory board).• Dependency parsing for recipe domain. Using the modern Deep Dependency Links: Resulting system was better than using Google dependency parser!• Link generation (identifying recipes sites to crawl).• Categorization of crawled pages.• Domain exploration (identifying capabilities of connected kitchen devices to be extracted automatically).• Recipe field extraction and verification from crawled data.• Creation of Data Science Workbench for automatic conversion of recipes.• Recipe terms identification (mention detection, phrases and their semantic types).• Syntactic clause detection.• Analysis of ingredients: ingredient name, quantity, measure, preparation.• Recipe flow inference (splitting of incremental ingredients, implicit container identification).• Coreference of ingredient mentions in recipe instructions to recipe ingredients.• Coreference of containers.• Semantic role labeling: Source and target container relations for cooking actions.• Review and correction of auto converted recipes (both [1] semantic layer, and [2] end-user facing recipe creation and editing system).• Identifying missing resources for recipes and adding to system (ingredients, cooking actions, secondary actions, appliances, settings, containers, tools, utensils, …): 10X increase of ingredient database!• Recipe taxonomy induction and update (1K+ categories).• Recipe categorization into taxonomy nodes.• Creating pristine recipes for demos to investors.• Creating comprehensive user manuals and annotation guidelines.

San Francisco, California, Us

Leading the Data Science team. Working on:+ Search and Discovery.+ Recommendations.+ Personalization.+ Advertizing.+ Optimization.+ Prediction.Using:+ ML: Machine Learning.+ NLP: Natural Language Processing.

Senior Manager, Local Search and Principal Research Scientist, Product Search.+ Launched Local Search beta version.+ Competitive analysis across Amazon and other Local Search Providers.+ Query understanding: Location extraction.+ Semantic Search & NLP (Natural Language Processing):++ Product type identification.++ Semantic types identification.+ Search results evaluation.+ Drove annotation efforts (both internally within Amazon and externally with an annotation vendor).++ Working with Legal team on Master Services Agreements and Work Orders.+ Product and Project management.+ Evaluation of startups for acquisition.

Redmond, Washington, Us

Leading Bing Local Search Data Measurement and Local Search Relevance Measurement.+ Local Search attribute accuracy measurement.+ sNDCG: Signed, Normalized, Cummulative Gain for Local Search relevance.+ Local Entity conflation/matching.+ Local Entity annotation and categorization, clustering and taxonomy induction.+ Local Entity popularity modeling.+ Cross-lingual Local Search.

Troy, Mi, Us

Leading the R&D team in the Web Intelligence Division. Team of chief architect, senior engineers, scientists, manager of linguistics team (data researchers, annotators/judges, outsourced labelers), interns, professors members of Technical Advisory Board (Stanford, Univ. of Pennsylvania, Univ. of Colorado) with end-to-end responsibility for content analysis (natural language processing, sentiment analysis and topic detection and categorization systems).• Discovery of topics of conversation:++ Topic modeling for clustering, topic tracking over time, new topic identification.++ Content Search (Information Retrieval): Specifying large queries of phrases (thousands vs. 30 terms in search engines) using non-recursive, context-free grammars.• Sentiment analysis:++ Machine learning techniques for mention detection, dependency parsing, coreference, meronymy. Targeting of sentiment expressions. Determining the semantic orientation of mentions and aggregating across all mentions in a coreference/meronymy chain.++ Sentiment corpus: 2 larger and more detailed than any other sentiment resource: Entities: mentions (16 semantic types|automotive, camera & consumer packaged goods (cpg) domains), coreference relations, meronymy relations; Sentiment expression, negation,intensifiers, comparisons; relation visualization; annotation; annotation guidelines, creating a license and working with the legal department.++ Chinese sentiment analysis (use of Amazon Mechanical Turk for creating machine learning training datasets).++ Structural analysis (dependency parsing); State-of-the-art accuracy and ecient decoder; managing CoNLL-style annotation for out-of-domain data.

Us

Managed R&D team (scientists, chief architect, senior engineers, manager of linguistic annotation team, outsourced spam annotation team, interns, university professors members of the Technical Advisory Board).• Foundational NLP architecture: Crawling, de-htmlizing, tokenization (statistical and heuristic; emoticons), language identification (word-based n-gram models), token normalization; sentence detection; storing, indexing, retrieval (Lucene, UIMA).• Clustering for automatic topic discovery (phrases generalized through WordNet hypernyms; k-means).• Proximity-based sentiment analysis.• Developed fast, state-of-the-art spam detector (based on url segmentation and web content.• Part-of-speech tagger (efficient, Hidden Markov Model with tag pruning, and data-driven, semi-supervised, sffix tying model for unknown words; Transformation-Based Learning (TBL); Support Vector Machines - SVMs).• Phrase detector/chunker (Support Vector Machines (Yamcha); de-lexicalized, pos n-gram context model; fast finite-state chunking; Transformation-based Learning (fn-TBL).• Social Network Analysis (SNA): tribe analysis system (comparisons of clusters of content from two sets of authors, visualization{guess system).• Demographic analysis (gender and age) based on user handles, URL segmentation, patterns and text categorization.• Supported sales, marketing and thought leadership efforts through client presentations, industry speaking events, and press interviews.• Evaluated third party analytic components for licensing by Umbria Inc.

Armonk, New York, Ny, Us

• Winner, U.S. Government (NIST-run) competition on text analytics:TERN (Time Expression Recognition & Normalization):– Used Robust Risk Minimization and Maximum Entropy techniques;– Developed a “conservative” tokenizer;– Used distributed environment for quick experimentation;– Built system for visualizing mentions.• E-commerce: Web-based dialog recommendation:– IAAI award (Innovative Applications of Artificial Intelligence) for deploying a live system.– System shown to significantly reduce the length of interaction in terms of time and number of clicks in finding products.– Integrated Decision Tree-based (constituency) parser.– Internal semantic representation, mapping to SQL, choosing follow-up questions, relaxation of constraints, database back-end.• Lead cross-team effort on entity extraction (mention detection) between Information Extraction, Question Answering, Finite-State Processing teams within IBM:– Awarded IBM division-level achievement.– Showed higher accuracy of combined mention detection system which uses machine learning module, question answering module, and finite state module.– Liaised with different teams to achieve a common code base and ran individual components.– Performed stream synchronization (via token alignment) and built combined system.• ACE (Automatic Content Extraction - entity detection & tracking):– DecisionTreess, RRM & MaxEnt.– Explored n-gram features combining: lexical stems, capitalization, POS, chunking (induced from Penn TreeBank first-level constituent labels), dependency-based information (using English Slot Grammar), WordNet, gazetteers, also streams which are the output of other classifiers (HMM) trained on other data (MUC, CoNLL).– Created rule-based coreference system.

Brighton, East Sussex, Gb

Robust parsing with wide-coverage grammars for English:– Designed and implemented an integrated environment for parsing and grammar development using a variant of Tree Adjoining Grammars; visualizing parse structures including feature structures and co-references.– Parsing through compilation to Linear Prioritized Multiset Grammars.– Developed and implemented algorithms for large-scale grammar development using a specialized language for describing non-monotonic inheritance hierarchies (DATR).– Compiled feature structures to flat terms for efficient unification during lookup and parsing.– Encoding semantic information.– Finite-state parser which uses precompilation techniques in order to speed up on-line processing.– Calculated compact form of expanded grammar from hierarchical encoding.– Precompiled sets of trees associated with the same word class (computationally intensive).– Built in a team effort the largest grammar of English with over 3500 constructions.