Zheng (John) Jiang AI apps and robotics demos

Field notes from AI application work

AI demos that run, leave logs, and can be checked by another engineer.

I build practical AI application work around Qwen VLM demos, FastAPI/vLLM APIs, batch detection, RealSense + SAM tracking, Jetson AGX Orin tests, and RAG/LoRA experiments. I keep inputs, logs, screenshots, and setup notes close to the work so another engineer can check it.

Open demo lab Read project notes GitHub Resume
Education
BSc (Honours) in Computing, The Hong Kong Polytechnic University
Best fit
Entry-level AI developer, GenAI developer, AI application engineer, software engineer
Core stack
Python, FastAPI, vLLM, PyTorch, Docker, Linux, React, Node.js, Nginx
Demo Lab Qwen, DeepSeek, tracking, chat, SaaS replay Projects Search by stack, project name, or evidence Preview Open focused screenshots and demo notes Experience Recent AI and infrastructure work
01

Perception

VLM inputs, RealSense color/depth streams, masks, boxes, and object-state output.

02

Reasoning

Prompt design, JSON parsing, RAG/LoRA experiments, and model API workflows.

03

Control Handoff

2D/3D anchors, pose hints, logs, CSV summaries, and notes another engineer can inspect.

04

Edge Testing

Jetson AGX Orin, JetPack 6.2, RK3588/RKNN, Docker/Linux, and deployment checks.

Working stack

Tools I have used in shipped or demo work

Python FastAPI vLLM PyTorch Hugging Face Qwen VLM Gradio Docker Linux Nginx React Node.js Firebase RealSense SAM 3 Jetson AGX Orin
6

project notes with source links or demo previews

3

AI workstreams: model APIs, multimodal detection, robotics perception

1

static Cloudflare Pages site, no embedded API keys, resume included

Run evidence

Screenshots from projects I started and inspected locally.

I keep the evidence visible because it is more useful than a polished claim. Some older projects run cleanly; some need dependency fixes or external services, which I note directly instead of hiding it.

Property Management System owner dashboard running locally
Local run Property Management System Express app running on localhost, owner login tested with session-backed dashboard.
RentConnect React application running locally
Local run RentConnect React app compiled after restoring a missing image asset; shown with real local page output.
E2EE Chat Docker Compose service layout terminal evidence
Terminal evidence E2EE Chat Docker/Nginx/MySQL/Flask deployment shape, with video evidence in the demo panel.
AI Finance Advisor local report workflow terminal evidence
Report workflow AI Finance Advisor Local report-generation workflow shown as a review-first AI draft system.

Robotics AI Demo Lab

Try small demos from the actual portfolio projects.

I kept this lab compact on purpose: each mode shows a runnable slice, a replay, or a constrained API-backed draft. Qwen uses DashScope through a server-side Pages Function; Property and Finance can use DeepSeek the same way, without exposing keys in browser code.

Reference-guided detection with recorded samples and an optional live Qwen3-VL call.

Qwen live VLM call Property staff draft Finance review draft Tracking object state Chat service flow RentConnect UI path

Endpoint status: recorded replay ready. Click Try live endpoint to call the server-side Qwen/VLM function.

Recorded reference input
Reference
Recorded search input
Search image
Recorded Qwen3-VL result
Recorded result
3 detections
7-8s recorded run time
JSON bbox output

Selected work

Project notes a recruiter or engineer can skim quickly

I kept these short on purpose. Each card gives the problem, what I built, and a piece of evidence. The demo buttons below update the interactive panel without leaving the page.

Showing 6 projects

Qwen3-VL detection output with bounding boxes

GenAI + computer vision

Qwen3-VL DuoImage Detector

Problem
Find objects in a search image that match a separate reference image.
Built
Python workflow for pairing, VLM prompting, JSON parsing, coordinate conversion, and rendered outputs.
Evidence
Reference image, search image, result image, JSON, CSV/XLSX summary path.
Source
Property Management System owner dashboard screenshot

Full-stack AI app

Property Management System

Problem
Turn tenant issues and staff updates into a visible workflow.
Built
Node.js and Socket.IO with login sessions, owner/staff dashboards, chat, and an AI reply path.
Evidence
Runnable prototype structure with end-to-end request flow.
Source

Robotics perception

RealSense + SAM 3 Tracking Prototype

Problem
Keep object state usable for robotics demos, not just visible on screen.
Built
Color/depth input notes, mask tracking, bbox, 2D/3D anchor fields, and pose cues.
Evidence
Canvas preview and structured object-state format shown in the demo panel.
E2EE Chat Docker Compose terminal evidence

Security + web

E2EE Chat Network Application

Problem
Design a chat app where transport, login, and message handling can be reviewed.
Built
Flask/MySQL service, Docker Compose, TLS/Nginx, OTP/MFA, and encryption design notes.
Evidence
Deployment notes and security flow documentation in the project folder.
Source
RentConnect local React app screenshot

React + Firebase

RentConnect Secure SaaS Platform

Problem
Build a real-estate workflow with accounts, listings, and a modern web UI.
Built
React app with Firebase Auth, Firestore, Material UI, Redux Toolkit, and payment wiring.
Evidence
Showcase image plus source folder with app structure.
Source
AI Finance Advisor terminal workflow evidence

AI workflow

AI Finance Advisor

Problem
Turn local financial context into a structured report draft.
Built
Python workflow that reads local inputs, calls an LLM API, and writes Markdown reports.
Evidence
Report skeleton and review-first workflow. It is not financial advice.
Source

Interactive preview

Demo panel

These are static previews, not live model inference. I use them to show inputs, outputs, and the small checks around each project.

How I work

A few notes that make the work less generic

I try to leave evidence behind.

For AI demos, I care about files and screenshots that let someone else verify the result. I keep input pairs, rendered outputs, logs, CSV summaries, timing notes and setup steps close to the work.

I use AI tools, but I still own the result.

My usual tools are Codex, Antigravity, Hermes OpenClaw and Coze. I use them for scaffolding, debugging, workflow drafts and documentation. The useful habit is checking assumptions, running the code, and trimming generated text until it sounds like something I would actually say.

I like messy prototype work.

My recent work often sits between model behavior, cameras, APIs, Linux boxes, and demo deadlines. I am comfortable making the first version work, then writing enough notes that the next person is not guessing.

Experience

Recent engineering work

Beijing Qianjue Technology - AI Developer Experience

Worked on robotics AI demos around Qwen-family models and model APIs. Day to day, I used FastAPI, Gradio, vLLM, RealSense/SAM tracking, batch detection scripts and edge AI tests on Jetson AGX Orin and RK3588/RKNN devices. The work covered setup, scripts, outputs, reports, timing checks and demo maintenance.

Dream Technology Solutions - Co-Founder and Infrastructure Lead

Handled client requirements and server setup for ESXi/RAID, Docker/Nginx, reverse proxies and tunneling. I also worked on BeamNG and Minecraft server deployment, mod testing, troubleshooting and handoff notes.

Microsoft and WeDragon - AI/Software Internships

Supported LLM deployment experiments, cloud server maintenance, AI course materials and Jupyter labs. I also helped with exams, project-site maintenance and documentation updates.

Contact

Open to AI developer and software engineering roles.

Best fit: LLM/VLM application engineering, Python/FastAPI services, robotics demo tooling, multimodal workflows, and full-stack prototypes that need practical delivery.

Email me GitHub portfolio LinkedIn