Bosheng (Daniel) Zhang
AI/ML Engineer · Data Scientist @ Similarweb
Designing and delivering end-to-end AI systems — from computer vision pipelines to LLM-powered platforms. Specialized in AI agents, RAG, and production AI solutions.
About Me
Hi, I'm Bosheng Zhang, but you can also call me Daniel. I'm an AI engineer with 5+ years of experience designing and delivering end-to-end AI systems — from problem scoping to production deployment.
I recently joined Similarweb in Shanghai as a Data Scientist, working with leading AI companies in China on the data side of model development — the gaps, bottlenecks, and evaluation work behind how teams build and benchmark their models.
Before that, four years at Vision HQ in Brisbane architecting AI for Australian local governments — large-scale computer vision for road and street monitoring, and an LLM-powered platform for automated analysis and reporting on GCP.
My strength is connecting the dots across the full solution lifecycle — data pipelines, model optimization, cloud infrastructure, API integration, and delivery that non-technical stakeholders can actually use. Outside of work I keep building in agentic AI, RAG architectures, and domain-specific LLM applications, and I run Chance Co. with a few engineer friends — we build custom web systems and data solutions for businesses on the side.
Experience
Data Scientist
Similarweb
AI & Software Engineer
Invespec
- Designed and built a CRM system for a certification and audit business, tracking client engagements across its certification service lines
- Explored and developed internal AI tooling to boost productivity across the team
AI/ML Engineer
Vision HQ
- Architected an AI-powered road monitoring system deployed across council road networks at scale, cutting survey costs by replacing dedicated survey-vehicle deployments with automated detection
- Designed and developed a street waste detection system generating city-wide Cleanliness Index scores, enabling data-driven cleanup dispatch and equitable resource allocation
- Built an LLM-powered IoT monitoring platform on GCP using function calling to orchestrate Elasticsearch data retrieval and anomaly detection, replacing manual daily telemetry review with automated reporting
Bachelor of Computer Science
University of Queensland
- Major in Machine Learning, with coursework in advanced algorithms, pattern recognition, and computer systems
- Completed course projects in medical image segmentation and reinforcement learning
Skills
AI / Machine Learning
Cloud & DevOps
Languages & Frameworks
Data & Tools
Featured Projects
Production AI/ML systems and selected research projects. Click any card to read the full case study.
AI-Powered Road Infrastructure Monitoring System
Computer vision system mounted on waste trucks for automated city-wide road defect detection, enabling proactive infrastructure maintenance.
AI-Powered Street Waste Detection & Cleanliness Index
Computer vision system that detects street-level waste and generates a city-wide Cleanliness Index, giving councils a data-driven view of urban cleanliness.
LLM-Powered IoT Device Monitoring Platform
Intelligent monitoring platform using LLM function calling to orchestrate multi-step telemetry investigation, cross-signal anomaly correlation, and automated report generation for IoT device fleets.
Academic & Research Projects
Skin Lesion Segmentation with Improved U-Net
Deep learning model for melanoma detection achieving 0.8+ Dice score on the ISIC dataset, using an improved U-Net architecture.
GreenMiles - Sustainable Transportation App
React Native mobile app incentivizing public transportation through gamification, carbon footprint tracking, and social connectivity.
LaserTank AI Solver
AI solvers for the LaserTank puzzle game using A* search, MDP planning, and reinforcement learning algorithms.
CANADARM Motion Planning
Probabilistic Roadmap-based motion planning for the ISS Canadarm2 robotic arm, navigating obstacles in constrained 2D workspace.
RUSHB Network Protocol Suite
Custom network protocol implementation with server, adapter, and switch components for reliable data transmission across networks.
Blog
Thoughts on AI, engineering, and building things.
I Built RAG for a Domain Chatbot, Then Deleted It: When You Don't Need Vector Search
A build log for a customer-facing construction documentation assistant. How I started with client-side RAG, shipped a server-side RAG, and finally deleted all of it in favor of full-context injection plus prefix caching — and the rule of thumb that tells you which one you actually need.
macOS VPN Split Tunneling: From 267ms to 28ms Without Touching Your VPN Config
How I built a client-side split tunneling system for macOS that routes local traffic direct and keeps international traffic on VPN. Covers IP-based route splitting, reversed DNS splitting (clean DNS via VPN tunnel + local DNS for CDN optimization), and a launchd auto-watch daemon.
Running Claude Code on DeepSeek V4: Same Tool, 7-89x Cheaper
How I built claude-ds — a drop-in wrapper that runs Claude Code with DeepSeek V4 as the backend. Covers the architecture, environment variable reverse-engineering, vision workarounds, and honest tradeoffs from daily use.
Let's Connect
I'm always open to discussing AI/ML projects, freelance development work, or just connecting with fellow engineers. Feel free to reach out!
ddaniel.zhang0413@gmail.com