Building a Self Service DBaaS for internal developing platform@Low Latency Data Store

- Designed, developed, tested, deployed, and maintained Django app generating Pandas DataFrame reports, showing usage and capacity of satellite internet services based on dates, markets, satellites as solo developer- Refactored codebase from procedural to object-oriented by designing system and class diagrams based on peer reviews and SOLID principles to enhance maintainability and scalability- Implemented error handling and optimized user data queries from BigQuery tables in endpoints built with FastAPI, improving slot time consumed by 50% and deploying to AppEngine to migrate APIs from Django app- Automated weekly report generation by writing shell scripts executing Django custom management commands to create reports and adding scripts to Cron to be run every week, reducing task time by 250%- Wrote Python scripts to develop data pipeline ingesting geospatial data of satellite beam polygons in KML files, converting them into CSVs and GeoJSONs, and loading them into BigQuery tables

- Maintained firmware detecting motion using WiFi waves by analyzing logs for debugging, building code into binary and porting it to OpenWRT for testing, and deploying firmware over-the-air to customer devices

Neuromodulation Project: Real-time Neural Decoding System from Neuron Firing Video for Prediction of Behavior (Advisor: Chen Rong, Shuvra S. Bhattacharyya)• Trained LSTM/GRU with online and transfer learning in TensorFlow to predict existence of fine movement from mice’s locomotor activity time-series data streamed from calcium imaging device in real-time (100ms inference)• Improved accuracy by 20% and F1 by 3% over offline batch learning

Publication: Resilience of Autonomous Vehicle Object Category Detection to Universal Adversarial Perturbations (IEEE IEMTRONICS 2021, Best Oral Presentation Award, Advisor: Mohammad Nayeem Teli)• Curated COCO17 dataset to five autonomous driving-related categories (people, car, stop sign, traffic light, truck), with each category dataset selected with images that Faster-RCNN detected at least one instance of that category as baseline• Implemented Projected Gradient Descent algorithm to compute an universal perturbation blinding Faster-RCNN in detecting target category when added to the dataset used for perturbation computation• Blinded 70% of images in 4 category subsets with perturbation norm less than 1/10 of image’s (Published in 2021 IEEE IEMTRONICS as co-author and awarded Best Oral Presentation)

Project: Chicken Farm Monitoring System• Wrote software detecting human annotation errors and converting label formats for Detectron2 object detection library • Trained Faster-RCNN and Mask-RCNN to replace YOLOv3 detector in DEEPSORT object tracker to improve tracking-by-detection

Course: Networks & Protocols 2 (ENTS641)• Revised and graded assignments on OSPF, IP subnetting, distance vector and link-state protocols given to 40 students

Publication: Lung Cancer Subtype Deep Learning Classifier based on 2D Joint Histograms of multi-modal CTs (IEIE 2018, Advisor: Taesup Moon)• Preprocessed 3D CT scans into 2D joint histograms to extract discriminative features and reduce training time• Trained VGG16 on histograms to classify ordered 1st and 2nd dominant subtypes out of 6 subtypes • Boosted accuracy by 3% from baseline predicting with an average probability distribution of training labels

[Projects]1. Cross merchandising with liquors in hypermarkets for increasing adult sparkling consumption• Wrote business proposal identifying business opportunities for cross-merchandising sparkling soft drinks with liquors in hypermarkets to increase early adulthood (age 25-34) consumption2. Monthly Sales Training Video Development to Replace Time-Consuming Sales Rally• Coordinated marketing and sales teams’ feedback with the production agency in developing sales training videos, distributed to +120 sales reps to align them with product knowledge and sales strategy 3. Sales Channel Development in Fast-Growing Upscale Foreign Restaurants• Installed Coke frames in 30+ stores to build customer relationship• Wrote sales manual on channel development in upscale restaurants, increasing revenue by 30% and delivering it to scale up to 500+ stores[Daily]• Gained insights into promotion depth vs. competitors in modern trade channels from monthly market research• Analyzed three-year data of convenience store promotion frequencies by brand/account to predict future trend

[Product]1. Trial feature development for up-selling your existing customers• Researched trends and competitors in language learning market to define requirements and features for free trial product• Calculated financial projections based on key variable assumptions to create business model • Defined KPI and its data flow in UI and tested with software development team to track and analyze user data 2. Product Mix Reassortment on New Product Launch• Analyzed sales volume and revenue of existing product mixes to decide new product mix including recently launched Japanese education product[Data Analysis]1. Customer Order Rate Turnaround in the Telesales Channel• Investigated conversion rate of sales reps by source of sales leads to optimize lead distribution by Excel pivot tables