Zehan (Henrick) Lin
Education
University of New South Wales (UNSW), Sydney
Feb 2024 - Jun 2026Bachelor of Computer Science (WES iGPA: 4.0/4.0, WAM: 83.5)
Relevant Courses: Web Front-End Programming (96), Software Engineering Fundamentals (96), Computer Systems Fundamentals (87), Programming Fundamentals (85), Data Structures and Algorithms (Distinction), Object-Oriented Programming (Distinction), Software Construction: Techniques and Tools (Distinction), Extended Security Engineering and Cyber Security, Database Systems
Work Experience
AI Full-Stack / Algorithm Engineer Intern – Cool AI
Dec 2025 - May 2026
- Dockerized multimodal perception models (audio / vision / speech), reducing the cost of switching between local and server environments and improving AI application delivery efficiency.
- Designed model inference APIs and service startup flows with FastAPI, supporting fast integration, validation, and business scenario adoption.
Projects:
Document Compliance Review Platform
FastAPI, React, Vite, LLM, RAG, Qdrant
- Built an end-to-end document compliance review platform with FastAPI + React/Vite, covering document parsing, clause-level and full-document checks, violation annotation, and review report generation.
- Designed and implemented a hybrid Rule Engine + LLM + Guardrail architecture, adding controllable constraints and output validation on top of AI generation to improve accuracy, stability, and explainability in complex compliance scenarios.
- Built a RAG review pipeline using Qdrant vector search and few-shot enhancement with similar violation cases to provide dynamic grounding for the model; also refactored prompt engineering to reduce average clause-level prompt length from 15,299 to 2,918.
- Designed a risk-tiered dynamic reasoning switch that enables deep reasoning only for high-risk clauses, reducing typical review time from 853s to 652s for an overall speedup of about 23.6%, reflecting technical decision-making and cost-benefit tradeoff awareness.
- Built a Human-in-the-Loop feedback loop that routes reviewer labels (correct / incorrect / missed) back into the vector case library, using human review to cover AI boundaries and continuously improve false positive and false negative behavior.
Full Stack Developer Intern – Graviti
Dec 2024 - Dec 2025
Projects:
AI Image/Video Generation Platform (www.diffus.me)
Vue 3, Nuxt 3, Tailwind CSS, MongoDB, FastAPI, SQLAlchemy, i18n
- Built the official website for an AI image/video generation platform using Vue 3 and Nuxt 3, and pushed the features to production.
- Completed the migration to MongoDB, including data cleaning, schema restructuring, and frontend consumption model design, improving page rendering and content organization efficiency.
- Implemented i18n internationalization, and used Nuxt SSR for SEO optimization to improve global accessibility, search engine indexing, and first-screen loading performance.
- Took features from requirement breakdown and solution design to implementation, using FastAPI + SQLAlchemy + MySQL to deliver comment features and close the loop from product idea to usable functionality.
Extracurricular Activities
Director of IT – UNSW Artificial Intelligence Society (AISoc)
Mar 2025 - Dec 2025
- Built a Discord meeting-to-task system with Next.js 15 + FastAPI + PostgreSQL, delivering the full-stack product from requirement understanding and system design to a runnable implementation.
- Designed a Discord Bot to capture voice streams, connecting audio collection, speech-to-text, and LLM task generation into an automatically running intelligent workflow.
- Used Prompt Engineering to extract structured tasks and perform semantic deduplication, while supporting recursive modeling and dynamic interaction for multi-level subtasks, creating a reusable AI workflow.
- Used React useTransition to optimize large state updates in complex task lists, reducing UI blocking and improving product usability under high-interaction workloads.
- Maintained and continuously updated the society website (https://unswaisoc.com/) to support daily operations and external presentation.
Projects
MockMate - AI Interview Training and Evaluation Platform
Jul 2025 - Present
Python, FastAPI, SQLAlchemy, SQLite, React, TypeScript, Vite, Zustand
- Independently delivered an AI mock interview product from frontend/backend development to production deployment, using React + FastAPI to implement the full loop of question generation -> follow-up -> answer -> AI evaluation -> record retention.
- Implemented SSE-based main-question and follow-up generation for both general interviews and resume-project technical drilling, including frontend incremental rendering, non-streaming fallback, DB flushing, and local draft recovery.
- Built resume parsing with strict JSON LLM extraction to convert uploaded resumes into evidence-backed project structures, enabling project-context technical questions grounded in real resume claims.
- Designed a structured AI evaluation pipeline that enforces strict JSON multi-dimensional scoring and evaluates complete follow-up conversations to improve parsability, consistency, and record reusability.
- Automated RSS-driven question production with APScheduler, using LLM relevance filtering before persistence to keep generated interview content focused on high-signal topics.
- Built a GitHub Actions CI/CD pipeline with automatic PR checks and SSH-based deployment after merge, demonstrating the ability to independently deliver a production-ready online product.
Awards & Competitions
•Ranked 9th across Australia and New Zealand (Top 10%)-AllUni Programming Competition Division B
Aug 2025Technical Skills
Programming Languages:
Python, TypeScript, JavaScript, Java, C, Rust, ShellFrontend and Full-Stack Development:
React, Next.js, Vue 3, Nuxt 3, Vite, Tailwind CSSBackend and Service Development:
FastAPI, SQLAlchemy, Express.js, RESTful API, SSEEngineering and Deployment:
Git, Linux, Docker, GitHub Actions, CI/CD, Shell ScriptingTesting:
Jest, JUnit, CypressDatabases and Retrieval:
PostgreSQL, MongoDB, MySQL, SQLite, Qdrant, LightRAGAI Coding:
Codex, Claude Code, Cursor