Hsiang-Shun Shih Email & Phone Number

Irvine, Ca, Us

Taoyuan City, Taiwan

• Fine-tune a Large Language Model(LLM) for Adlink's products using QLoRA, aiming to create a Document-Based Question Answering System with the state-of-the-art open source Mistral model. 720 datasheets for company products are crawled, from which we crafted a QA dataset consisting of 300 pairs. Tests confirm the enhanced Mistral model proficiently handles queries about product specifications, accurately responding to even rephrased questions not present in the dataset. • Implement… Show more • Fine-tune a Large Language Model(LLM) for Adlink's products using QLoRA, aiming to create a Document-Based Question Answering System with the state-of-the-art open source Mistral model. 720 datasheets for company products are crawled, from which we crafted a QA dataset consisting of 300 pairs. Tests confirm the enhanced Mistral model proficiently handles queries about product specifications, accurately responding to even rephrased questions not present in the dataset. • Implement Retrieval Augmented Generation(RAG) and Model Branching for LLM, integrating the Mistral model with the LangChain framework for Adlink products comparisons and porting user queries to appropriate models. Show less

Irvine, California, United States

• Supervised by Professor Phillip Sheu, research topic mainly lies in graph theory, including random graph sampling which maintains shortest path properties, influence maximization problem in a colored network, s• Aimed to derive a statistical sampling that given a set of shortest path property and evenly sample all the possible graphs. • Derived a guaranteed 1 - 1/e prediction lower bound of greedy algorithm of influence maximization, proved monotonicity and sub-modularity properties… Show more • Supervised by Professor Phillip Sheu, research topic mainly lies in graph theory, including random graph sampling which maintains shortest path properties, influence maximization problem in a colored network, s• Aimed to derive a statistical sampling that given a set of shortest path property and evenly sample all the possible graphs. • Derived a guaranteed 1 - 1/e prediction lower bound of greedy algorithm of influence maximization, proved monotonicity and sub-modularity properties for the influence function of a new influence diffusion model. Show less

Redwood City, California, United States

• Aimed to achieve Zero-downtime Orchestration during service mesh upgrading for Kubernetes. Utilize service mesh metadata, which includes artifacts compatible with connection descriptions, to perform compatibility checks.• Implemented a Golang solver to check the compatibility result and generate a upgrading plan by DFS heuristics regarding Manhattan distance, which can traverse possible artifact-to-service assignment in the service assignment space. Also, I have implement a CUE analyzer… Show more • Aimed to achieve Zero-downtime Orchestration during service mesh upgrading for Kubernetes. Utilize service mesh metadata, which includes artifacts compatible with connection descriptions, to perform compatibility checks.• Implemented a Golang solver to check the compatibility result and generate a upgrading plan by DFS heuristics regarding Manhattan distance, which can traverse possible artifact-to-service assignment in the service assignment space. Also, I have implement a CUE analyzer in order to check the validity of the service mesh upgrading plan. Show less

Taichung City, Taiwan

• Developed a Full Stack web project which aims to generate Rank Margining Tool(RMT) report and display it though web browser based on the extraction results from the back-end server command line tool is implemented. Flask is used to setup the back-end server, Node.js and JQuery are used for the front-end.

Taipei, Taipei City, Taiwan

• A Ministry of Science and Technology program, supervised by Professor Mao-Chao Lin. Implemented different A* algorithm optimization method, stimulated the decoding process and the bit error ratio through C++.

Doctor Of Philosophy - Phd, Computer Engineering

Bachelor Of Science - Bs, Electrical Engineering