Building deep learning language models for PII (personal info), NER and intent classification.𝐏𝐈𝐈:• Designed & implemented PII inference chunking pipeline architecture to support entity masking on largedocuments for Generative AI LLMs (batching, DJL routing, concurrency)𝐈𝐧𝐭𝐞𝐧𝐭 𝐂𝐥𝐚𝐬𝐬𝐢𝐟𝐢𝐜𝐚𝐭𝐢𝐨𝐧:• Launched a new intent product for chatbot application, supporting all stages of product development: market research, technical NLP model research, developing prediction pipeline, training framework through internal SDKs, model evaluation, customer feedback𝐍𝐄𝐑:• Architected prediction pipeline redesign for higher throughput/lower latency • Led efforts of multilingual expansion to 40+ languages and 10+ entities for contextual chatbot applications. • Implemented model training, augmentation, async predictions, normalization, GRPC contracts

• Product attribute extraction from text data for Jetlore• User-Product recommendations

𝗣𝗮𝘆𝗣𝗮𝗹 𝗖𝗼𝗺𝗺𝘂𝗻𝗶𝘁𝘆 𝗨𝗻𝘀𝘂𝗽𝗲𝗿𝘃𝗶𝘀𝗲𝗱 𝗣𝗼𝘀𝘁 𝗖𝗹𝗮𝘀𝘀𝗶𝗳𝗶𝗰𝗮𝘁𝗶𝗼𝗻• Predict category of user’s post to mitigate cluttering, false categorization and auto-moderation.• Scripted to fetch 46,000 public posts spanning 7 categories• Used Doc2Vec/autoencoders, researched hierarchical NLP models & used dimensionality reduction techniques (t-SNE, PCA) to cluster using K-means with silhouette scoring.𝗦𝗲𝗻𝘁𝗶𝗺𝗲𝗻𝘁𝗮𝗹 𝗔𝗻𝗮𝗹𝘆𝘀𝗶𝘀• Model intended to be used for identifying sentiment of customer messages to route requests accordingly• Convolutional/Bi-LSTM models with GloVe embeddings--achieved 87% accuracy• Used K-fold cross-validation, batch training & precision/recall metrics to identity false pos/negs

• Designed server capacity analysis dashboard in Node.js for Oracle Connection Cache (OCC) to detect & monitor anomalies in data, automate server capacity projections & visualize pool layout.• Host traffic analysis/automation b/w data sources• Disaster recovery compute parity with LIVE site• Maintained Hibernate-orm for OCC test measures

𝗧𝗿𝗼𝗻 𝗪𝗮𝗹𝗹𝗲𝘁 𝗖𝗼𝗺𝗽𝗲𝘁𝗶𝘁𝗶𝗼𝗻 (𝟮𝗻𝗱 𝗽𝗹𝗮𝗰𝗲 𝗶𝗻𝘁𝗲𝗿𝗻𝗮𝘁𝗶𝗼𝗻𝗮𝗹)• Designed hot, cold, watch-only wallets to send crypto transactions, view chain performance/transaction data with automatic selection of proof of stake nodes based on smallest latency. • GRPC, Java Spring API backend𝗧𝗿𝗼𝗻 𝗗𝗮𝗽𝗽 𝗜𝗻𝘁𝗲𝗿𝗻𝗮𝘁𝗶𝗼𝗻𝗮𝗹 𝗖𝗼𝗺𝗽𝗲𝘁𝗶𝘁𝗶𝗼𝗻 𝗙𝗶𝗻𝗮𝗹𝗶𝘀𝘁 (𝗗𝗲𝗺𝗼: https://bit.ly/2HKst8T)• Built crypto name service (DNS-like human readable addresses) for Tron • Wrote decentralized encrypted smart contracts on Tron network for updating aliases, permissions, tags for base58 addresses• Created responsive React/Semantic frontend with Node.js backend to send smart contract calls

𝗖𝗜𝗙𝗔𝗥-𝟭𝟬𝟬 𝗢𝗯𝗷𝗲𝗰𝘁 𝗖𝗹𝗮𝘀𝘀𝗶𝗳𝗶𝗰𝗮𝘁𝗶𝗼𝗻• Conv net with softmax classifier to denoise, normalize & classify colored images of objects• Experimented with feature maps, network structure, feature visualization of each layer𝗖𝗵𝗮𝗿𝗮𝗰𝘁𝗲𝗿-𝗹𝗲𝘃𝗲𝗹 𝗥𝗡𝗡 𝗼𝗻 𝗡𝗼𝘃𝗲𝗹𝘀/𝗦𝗽𝗲𝗲𝗰𝗵𝗲𝘀• RNN model to predict character sequences with largest probability• Improved performance by adding randomness to data & softmax via simulated annealing𝗠𝗡𝗜𝗦𝗧 𝗗𝗶𝗴𝗶𝘁 𝗖𝗹𝗮𝘀𝘀𝗶𝗳𝗶𝗰𝗮𝘁𝗶𝗼𝗻• Multi-layer network (w/ back propagation) from scratch using logisitic/softmax regression• Utilized different gradient descent techniques to better recognize digits

Successfully found web exploits in Facebook, Buzzfeed, and Stage32.com𝗙𝗮𝗰𝗲𝗯𝗼𝗼𝗸 [Medium post below]• Found a large scale security bug that let you gain complete access to multiple accounts at once.𝗕𝘂𝘇𝘇𝗳𝗲𝗲𝗱• Wrote multiple spiders: Mechanize, Selenium, PhantomJS, BeautifulSoup, Rotating IPs, Google Computing infrastructure, Multithreading (923+ http requests per second).• Complete account access (stealing session keys/brute force password) and DDoS memory attack.𝗦𝘁𝗮𝗴𝗲𝟯𝟮• Complete database takeover.

𝗔𝗰𝗲𝗯𝗼𝗼𝗸𝘀: iOS app for students to buy/sell crowdsourced college textbooks with automatic bidding!• Used web scraping to collect textbook data & processed it into large data sets in order to create efficient & automate algorithmic flow• Python scripts to automate the process of handling backend data and updating andmaintain database systems.𝗦𝗽𝗲𝗰𝗮: Developed an iOS/Android social networking app for sharing ephemeral images/videos!• Provides streaming and direct popularity to trending events and people.• MYSQL database using PHP and coded apps in Objective-C, Java and HTML web server.• Built attractive UI; Anonymous & engaging platform; Faster methods to taking pictures & enhancing them.

Created automated graphics driver tools for DirectX12 and working with large JSON data sets to make testing faster for developers. • Generated C++ code using python with the purpose serving as a beautifier, scripts for Microsoft's 3D graphics API (xBox)• Executed code review processes through Perforce & wrote custom-built source code management tools for Perforce• Wrote Python plugins to help with reconfiguring Intel server architecture development; product shift to remote debugging interface.

Designed and coded a fully functional educational software called "Eternity" from scratch to be implemented in the local school district. Pitched & productionized the software for public implementation AWARDS:3rd in Math & Comp. Science at Science & Engineering FairUS Army bronze medalNaval Force awardAir Force award

Public Relations & Media Director:Develop and execute an integrated public relations strategy to maintain a positive and attractive perception of the academy with key audiences Collaborated with mentors of other schools around CA, Microsoft & Google partners at the California Partnership Academy Conference to encourage and coach other academies of the secrets of building institutions on student leadership and service learning projectsLeader of Membership Committee:Supervised and managed the attendance of academy's students to monthly meetingsIncreased student participation to the meetings/events