• Studied the wheeled robot chassis documentation and controlled it through its ROS interface for forward, backward and turning motion.
• Deployed and benchmarked several YOLO image-recognition options on an NVIDIA Jetson Nano to find a configuration that suited the hardware.
• Built a working human-following prototype that used bounding-box size as a distance proxy and bounding-box position to decide when to stop, turn or move forward (obstacle avoidance was out of scope).
• Studied the company positioning-camera documentation, including how to obtain 3D coordinate data over the serial interface and image data over USB.
• Worked on a two-tracker demo that sent positioning data back to a host computer over a wireless link so two devices could be displayed on one machine; device poses were aligned by starting both from the same initial position and orientation.
• Deployed a YOLO-based vision model to a new, sparsely documented development board with a restricted model/runtime format: converted the model (ONNX) and traced unsupported-operator fallback that pushed part of the workload onto the CPU.
• Supported the technical marketing side: SEO keyword and search-volume research, account setup (Discord / Instagram / YouTube), demo filming, and video post-production.

• Contributed to a Scrum-based team building an authenticated, map-based web product with school search, filtering, registration and role-based workflows.
• Took Product Owner and QA responsibilities across different sprints.
• Supported sprint planning and backlog preparation, and helped document requirements, acceptance criteria and showcase outcomes.

Climate Claim Fact-Checking Pipeline

Hybrid retrieval + re-ranking + stance classification over a 1.2M-passage corpus

• Fused TF-IDF sparse retrieval with all-MiniLM-L6-v2 dense retrieval using Reciprocal Rank Fusion, reaching a 0.908 any-hit retrieval rate at a top-500 candidate pool.
• Fine-tuned climatebert/distilroberta-base-climate-f as a binary selector to re-rank the pool down to the top 150 passages.

Stock Forum Summarization & Sentiment Analysis System

A multi-agent LLM pipeline that turns noisy forum chatter into structured stock signals

• Designed a LangChain multi-agent pipeline: a low-cost filter agent removes off-topic chat, a second agent checks whether a company name is really being used as a stock reference, and an analysis agent returns structured output with the stock, the reasoning and the sentiment.
• Found experimentally that asking the analysis agent for a written rationale alongside its conclusion produced more reliable outputs; structured rationale prompting is now part of the system design.

splitB — iOS Bill-Splitting App

SwiftUI group-expense app with a network-flow settlement optimiser, released on the App Store

• Implemented a network-flow-based settlement algorithm that restructures inter-member debt into a lower-cost repayment path instead of preserving the original pairwise debts.
• Built both persistence layers: Core Data for local data and Firebase Firestore for shared groups, with invite codes and real-time group updates.

Selected results: Deep Learning 84 · HD · Mobile Application Development 90 · HD · Introduction to Programming 87 · HD · Programming Fundamentals in Java 83 · HD · Japanese Introductory 2 82 · HD

Programming

Python · Java · JavaScript · TypeScript · Swift / SwiftUI

AI / Machine Learning

Machine Learning · Deep Learning · NLP · Computer Vision · Transformers · BM25 · Dense Retrieval · Cross-Encoder Re-ranking · Sentiment Analysis · OCR

Backend / Data

FastAPI · SQLModel / SQLAlchemy · SQLite · REST APIs · Data-processing pipelines

Robotics / Systems

ROS2 · Jetson Nano · Stereo vision · Linux · Hardware–software integration

Engineering tools

Git · Docker · Pytest · Playwright · Nginx · Certbot · ChatGPT Codex