Bohan Wu Email & Phone Number

Pittsburgh, Pa, Us

– Contributed to the development of LanTai, a software for automated text revision based on handwritten proofreading annotations.– Developed Flask APIs for YOLO and OCR model inference. Managed concurrency by implementing a task queue and building a scheduler API to control the transmission of task requests sent to sub-task APIs.– Deployed the application on AliCloud, containerized the application using Gunicorn and Docker. Implemented loggers to monitor API activity.– Developed and trained a DAMO-YOLO-based model to identify proofreading annotations in text page images, optimizing the model’s size and configuration, improving average precision from 0.45 to 0.84.– Enhanced prediction accuracy by 10% by incorporating standardized hand-written annotations into real-world publisher training data. Boosted model robustness by adding underperforming data and augmented data.

– Overview: A conversation-based speaker diarization module capable of separating speeches from different speakers and extracting dialogues specifically from the conversation with a target speech based on the context and timing.– Built data sets derived from AMI Corpus and ASR-RAMC-BigCCSC containing audio mixtures, individual speeches, timestamps and text dialogues.– Developed the summary layer that utilizes FiLM layers to fuze audio vectors and audio embeddings, and contributed to building the training pipeline.– Developed a pipeline using ESPnet to parse .WAV-format audio data and generate large-scale voice mixtures.– Replicated various state-of-the-art speech separation models such as TF-GridNet and Voice Activity Projection (VAP).

– Overview: A multi-agent systems study through a web-based game where the human's objective is to locate cursor positions that give the minimal cost values, with interference from the machine that plays by inputting values retrieved from a gradient descent algorithm.– Implemented codes for large-scale data collection, server hosting, data cleansing and parsing, and data visualization.– Produced heatmaps and box-and-wheskers figures from input data to demonstrate human actions underscenarios with different machine adaptation rates, and examine the input’s approximations to differentequilibria.– Showed adaptive machine-learners’ capability to steer the outcome of the game to different game-theoretic equilibrium.

– Contributed to the development of ModelFarm, an AI algorithm development platform that encompasses model training, model validation, model deployment, and more for more efficient algorithm development.– Developed model deployment and training methods in JavaScript, including invoking algorithms and retrieving results, generating performance evaluation results for algorithms, reconfiguring algorithms using accepted arguments, etc.– Implemented test scripts for the application’s functionalities to run model training and inference.

– Contributed to the development of SharkStore, an enterprise data service application. Developed test scripts to verify transmission correctness, check access permissions, and identify transmission issues during the process.– Developed Java logics that enable data transfer between clients and the storage for back-end APIs.

Master Of Science - Ms, Electrical And Computer Engineering

Bachelor Of Science - Bs, Computer Science