M.Sc. Data Science @ FAU · Computer Vision · Software Engineering · Embedded AI
Based in Nürnberg, Germany, I connect computer vision, machine learning and software engineering to turn ideas into working systems. I enjoy building the whole pipeline—from experiments and evaluation to interfaces, APIs and hardware integration.
Portfolio · CV · LinkedIn · Email
- Pursuing an M.Sc. in Data Science at Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU).
- Background in Computer Science and Technology at China University of Geosciences (Wuhan).
- Interested in computer vision, AI applications and hardware–software integration.
- I like understanding why models work—and turning them into useful applications and devices.
| Area | Technologies & interests |
|---|---|
| AI & computer vision | PyTorch, scikit-learn, OpenCV; object detection, image restoration, feature extraction |
| Web applications | Python, JavaScript, React, FastAPI, REST APIs, SSE, Gradio |
| Data & retrieval | NumPy, pandas, SQL, SQLite, FAISS, RAG, Streamlit |
| Embedded systems | C / C++, ESP32-S3, FreeRTOS, ESP-DL, camera interfacing, GPIO |
| Development & testing | Git, pytest |
A modular vision toolkit covering 3D geometry, RAW / HDR imaging, image retrieval, object detection and face analysis, exposed through one Python package, a unified CLI and a Streamlit studio.
- Robust plane fitting with RANSAC, MLESAC and Preemptive RANSAC, followed by PCA-based 3D box measurement.
- RAW demosaicing, camera-response calibration and HDR tone mapping; image retrieval with RootSIFT → VLAD → Exemplar SVM.
- Selective Search + HOG / SVM detection, open-set face recognition and unknown-identity rejection, with reproducible benchmarks.
Stack: Python · Streamlit · OpenCV · scikit-learn
A Munich hackathon prototype connecting map-based discovery, reservations, driver and operator dashboards, and ECURIOUS parking-lock integration.
- React / Leaflet interfaces and a FastAPI / SQLite backend for accounts, vehicles, reservations, waitlisting and reallocation.
- Official REST / SSE integration with plate-based check-in, event deduplication, state synchronization and failure recovery.
- Separate Mock and Live modes; charging, cost and revenue figures are simulated or estimated. Physical hardware validation remains pending as documented on the portfolio.
Stack: React · Leaflet · FastAPI · SQLite · REST · SSE
On-device vision firmware designed for person, bicycle and car detection, with ROI tracking, graded buzzer alerts and a Wi-Fi dashboard.
- Integrates the official YOLO11n COCO 320 INT8 model through ESP-DL for inference on ESP32-S3.
- A bounded FreeRTOS pipeline connects camera capture, inference, tracking and alerts.
- Includes model SHA and tensor-contract checks, host tests and guarded flashing; traffic-cone detection is an extension in progress.
Status: In development. Firmware builds pass; on-board hardware testing is pending. No measured FPS or accuracy is claimed. A research prototype, not a navigation-safety product.
Stack: C++ · ESP32-S3 · ESP-DL · FreeRTOS · YOLO11n INT8
A Streamlit application for document RAG, agentic retrieval, paper reading and tabular analysis, with English, Chinese, German and French interfaces.
- Document upload, chunking, embeddings and FAISS retrieval; an agentic pipeline adds routing, grading, query rewriting, generation and verification.
- PDF text and figure extraction, structured paper summaries, and retrieval-based or full-text Q&A.
- CSV / Excel cleaning, natural-language analysis, Plotly charts, multi-table joins and scikit-learn baseline models.
Stack: Python · Streamlit · FAISS · DeepSeek · Plotly · scikit-learn
EV Charging — First Prototype
The original EV charging repository validates the reservation-to-release business flow using an explicit Mock API.
- Separate React frontend, FastAPI application backend and Mock Server.
- Reservation conflicts, cancellation, role permissions, automatic check-in, incorrect plates, No-Show handling and blocked-bay alerts.
- The v2 project extends this architecture with official parking-lock REST / SSE adapters.
The original ECURIOUS API is a fictional hackathon mock; charging and revenue figures are simulations.
Bachelor's thesis: Prior-enhanced image restoration · December 2024 – June 2025
Designed PriAda-U-Net for dehazing and GRS-U-Net for deraining, integrating dark-channel and gradient-map priors. Evaluated adaptive loss weighting through comparative experiments and ablation studies.
Reported dehazing experiment: PSNR 34.13 dB · SSIM 0.8325. See my CV for context.
| Period | Role | Institution / organization |
|---|---|---|
| Oct 2025 – Present | M.Sc. Data Science | Friedrich-Alexander-Universität Erlangen-Nürnberg |
| Sep 2024 – Mar 2025 | Software Development Intern | Guanghong · Retail business |
| Sep 2021 – Jun 2025 | B.Sc. Computer Science and Technology | China University of Geosciences (Wuhan) |
During my internship, I independently developed and deployed a WeChat Mini Program for product browsing and online orders, with a Python backend and AI customer support, and managed testing and maintenance.
Happy to talk about computer vision, AI applications and software development.
- Email: yingchao.li@fau.de
- Portfolio: yingchao-li.github.io
- LinkedIn: Yingchao Li
Understand the principles. Build the whole system.