Yufei Wang

Software Engineer Internship at Feel

Feel is a new type of social media application geared towards supporting its users’ emotional wellbeing. On Feel, users can log and track their emotions to keep a pulse on how they are feeling. I involed in the chatbot development where users can talk with for additional support. Specifically, the chatbot can act as an empathetic listening ear, ask meaningful questions, recommend resources, comfort users, etc.

I firstly started by designing rigid conditions to decide wether the bot would act as a friend or a therapist using intent analysis and sentiment analysis powered by LangChain. Then I optimzed the by function calling chatbot using tools parameter to specify additional tools or features that should be enabled or utilized during the completion (or response generation) process.

The bot can also memorize what users said a long time ago, by vectorizing the conversations and storing in the PGVector concurrently while the bot is answering the user's questions.

Suggested by my manager, I further optimized the bot using Semantic Kernel developed by Microsoft, which involved developments of plugins to lead to more efficient and responsive applications.

I also created a Python script to evaluate the chatbot's performance on the test set using the BLEU metric and GPT-4o scoring to for better hyperparameter fine-tuning and prompt engineering.

Open-source Software Development

Fine-tuned LLaMa using Multi_lora on a code dataset, and added update parameters to base model during inference time.

Published the fine-tuned model to Huggingface. Model card

Test and optimized hyperparameters for MixLoRA (Mix-of-Experts) on the classification task. Repository

AI-driven recommendation assistant

The internship focused on developing a recommendation assistant for agents of Ping’an of China, one of the largest insurance companies in China, utilizing the open-source Large Language Model, ChatGLM. The recommendation process involved a four-step reasoning procedure designed to ensure a personalized and effective user experience. Additionally, the system can also provide recommendation rationales, helping users comprehend the reasons behind the recommendations it offered. For instance,

If a user requested recommendations for medical insurance for a 61-year-old individual, the system would respond as follows: (1/4) Recognizing the need for products recommendation and I am proceeding to offer suggestions. (2/4) Identifying various product options, such as "Hospital Insurance" and "Worry-Free Medical Care."." (3/4) Based on your description, "Worry-Free Medical Care" is more suitable for you. (4/4) Here are the reasons: The insurance age for this insurance is 18-65 years old, and the coverage covers 120 diseases, etc..

My responsibilities included data preprocessing, where I analyzed and cleaned the raw JSON data, and extracted keywords related to product characteristics using an IDF file. Due to lack of user queries for testing, I collaborated with the product manager team to build 10 test set consisting of 10 queries across 3 real-word scenarios: elderly individuals, frequent business travelers, and children. Furthermore, we opted to evaluate the model using two key metrics: "Helpfulness" to ensure that the model provides useful information, and "Honestness" to guarantee that the model refrains from fabricating information or misleading users.

I also engaged in prompt engineering by experimenting with different prompts, such as defined roles (e.g., assuming you are an excellent and professional account manager), specified scenes (e.g., explaining recommended products to users), word limits (e.g., 200 words), delimiters, incorporating non-linguistic context (e.g., marked with colons), and style induction(e.g., introduce friendly).

Using proper prompts, the system can achieve a 10/10 score in terms of Honestness.

Noteworthy Projects

E-commerce product flash sale system based on Golang

  • The characteristics of the flash sale system is that a lot of people would flocked in trying to place an order at the same time. So the system must have high concurrency to handle large traffic and also data consistency is really important.
  • The basic architecture is that when uservisit the page and send requests to web server, and the server would query data from the database. After getting the data, the web server would render the page to user, which involves product, user and order management. However, there are several problems with this architecture:
  • 1. Web server would need to handle lots of request and guarantee secure access, which would put a lot of stress to the web server. 2. Querying the database for every user request results in low performance. 3. Simply interacting between server and database during high traffic may not ensure accurate inventory management, potentially leading to overselling products beyond available quantities.
  • Thus I optimized the system's concurrency using several techniques: Deployed the front-end page using AWS CloudFront for CDN optimization.
  • Designed a distributed authentication module to filter traffic effectively using AES-encrypted cookies. I also implemented a consistent hashing algorithm to store user last login times, controlling access rates.
  • Utilized RabbitMQ to manage database access flow and process messages asynchronously.
  • Created a module that achieved over 15k QPS on a single machine deployed on AWS EC2.
  • Utilizing these techniques, the system can operate concurrently and handle high loads effectively

    • Online Dating Web Application

  • I developed an online dating website where you are free to make friends and share posts. You can visit it here, but note that I will charge for every messages you sent. It's hard to find real love afterall:)
  • I created features for sending smile, friends requests and messaging using Node.js, Bootstrap and MongoDB. And I used Stripe Checkout handling messages charges.
  • I also implemented post sharing with commenting and thumbs-up functionality, storing images in AWS S3 using AJAX.
  • When a new user is registered, an automatic welcome messages will be sent using socket and message notifications will be created at the header of the website using jQuery.

    • YouTube Clone

  • Built a website for users to view videos and upload videos when authenticated with scalability.
  • Created a video processing service to transcode videos using Express, TypeScipt and FFmpeg. Stored videos in Google Cloud Storage and associated metadata in Cloud Firestore. Used Cloud Pub/Sub for message communization.
  • Built the web client with React and Next.js; Showed videos by retrieving URLs from buckets, leveraging Firebase Functions.
  • Deployed services in a containerized environment with Docker and GCP Cloud Run.

    • HDFS Emulation

  • Emulating the Hadoop Distributed File System (HDFS) by self-designing its main tasks. The frontend page was implemented via Vue.js and the backend server was built using python and Flask.
  • Firebase was used for the database to store NameNode and DataNodes. Like HDFS, the NameNode has two main sections, iNodeSection to store inumber of each file and dirSection to explore directory structure.
  • The system can also accept shell commands from the terminal the sockets were used for communication.

    • Click Through Rate Prediction

  • Advertising analysis of predicting whether a mobile ad will be clicked using Wide&Deep model and its variations Deep&Cross model and DeepFM model.
  • The project was built using python and PyTorch and Spark SQL for big data OLAP.

    • Movie Recommendation Engine Development

  • Movie Recommendation System based on the matrix factorization method to recommend top k movies to a user by analyzing his/her reviewing history. Additionally, the system can discover similar movies within the dataset using by utilizing the movie factor matrix and computing Euclidean and cosine distance of the movie factors.
  • The project was implemented by Spark ALS and Spark SQL

    • LA Crime Analysis

  • Spatial and time analysis of crimes for a 10-year dataset of reported crimes from LA government based on OLAP and model building.
  • The project was implemented by Spark SQL, Spark Dataframe and Spark ML.

    • Stock Price Prediction

  • Stock price prediction based on LSTM model.

    • Other Projects

    Customer Churn Prediction

    Identify customers who are likely to churn in the future and analyze top factors that influence user retention.

    • Details

    Topic Modeling

    Cluster customer reviews into groups and discover the latent semantic structures.

    • Details

    Gas Meter Reading Recognition

    Reading recognition based on digital image processing and neural networks.

    • Details

    Face Recognition

    Face recognition in real-time video streams.

    • Details

    Fraud Detection

    Predict the probability of the first transaction of a new user is fraudulent.

    • Details

    Daily Movie View Time Prediction

    Predict how much a movie is consumed by the audiences per day based on Amazon Prime Video Dataset.

    • Details

    Car classification

    Car classification based on Stanford Car Dataset.

    • Details

    Skills

    I have been actively engaged in programming since 2018, consistently immersing myself in the realm of technology.

    Below are some of the tools and analytical skills with which I am proficient.

    Programming language

    Golang, C++, Java, C

    Python, SQL

    HTML/CSS, Javascript, TypeScipt

    Front-end / Back-end

    React, Redux, Vue, Next.js

    Node.js, Express, Flask

    FastAPI, Spring

    Socket, Websocket

    Development tools

    Git, Linux

    AWS, GCP

    Docker, Kubernetes

    Big Data Analysis & Management

    Spark (Spark SQL, Spark ML, Spark Dataframe, RDD)

    Hadoop

    PyTroch, Tensorflow

    MySQL, PostgreSQL

    MongoDB, Firebase, Amazon DynamoDB, HBase

    Deep Learning

    Convolutional Neural Networks (CNN)

    Recurrent Neural Networks (RNN)

    Long Short Term Memory (LSTM)

    Transformer

    Transfer Learning

    Large Language Models(LLMs)

    Embeddings

    Word2Vec

    Milvus

    FAISS

    Fast Map

    Data Visulization

    Tableau

    Python (Seaborn, Matplotlib, etc.)

    Excel

    Supervised Learning

    Linear Regreesion

    Logistic Regreesion

    K-Nearest Neighbors (KNN)

    Linear Discriminant Analysis

    Support Vector Machines (SVM)

    Random Forest

    XGBoost, LightGLM

    Unsupervised Learning

    Principal Component Analysis (PCA)

    K Means

    Gaussian Mixture Models (GMM)

    Optimization Techniques

    Local Search

    Dynamic Programming

    Constraint Satisfaction Problems (CSPs)

    Heuristic Seach

    Job Scheduling Problem

    Graph and Netwrok

    Bayesian Networks (BNs)

    Hidden Markov Models (HMMs)

    Complex Network

    Gephi

    Recommendation System

    Collaborative Filtering

    Matrix Factorization

    Feature Interaction (FM, FFM, PNN)

    GBDT + LR

    Wide & Deep, Deep & Cross

    AutoRec, Nerual CF, Deep Crossing

    DeepFM, NFM, AFM, FNN

    DIN, DIEN

    xDeepFM

    Get in touch

    My inbox is always open. Whether you have a question or just want to say hello, feel free to mail me!

    Mail Me