University of Ottawa · ECE
Current work moves through AI systems, sensing pipelines, and hardware-aware engineering rather than staying only at the UI or product layer.
Kair Wang · technical archive
Kair Wang
I build systems that sit between models, sensor data, and the hardware they depend on.
Former software engineer. Now an ECE master's student at the University of Ottawa, working where sensing pipelines, machine learning, and usable software meet.
The shortest accurate summary: ECE, IEEE, AI water systems, and a longer arc toward semiconductor-aware engineering.
Research log
The throughline is not “product thinking.” It is systems work moving closer to the metal.
This replaces the usual slogan section. The useful story is the actual track: software engineering, ECE graduate work, IEEE papers, and deeper semiconductor-facing curiosity.
AI water resource systems
The research thread is about tying data collection, modeling, and decision support into one system that can survive real operational use.
Research that stays accountable to reality
The second IEEE paper reinforced the same standard: the interesting part is not a model in isolation, but whether the full loop holds up when someone has to use it.
Silicon-based III-V, heterogeneous integration, lower-layer constraints
That longer arc matters because it points toward the part of engineering where semiconductor realities start shaping software decisions.
Featured system
Revo, stripped down to the moving parts.
I still keep one shipped product in the foreground, but the page shows its request flow, state handling, and admin control surface instead of retail copy.
Shipped product, read as architecture
A recommerce system is only interesting when the happy path ends.
The public storefront is only the surface. The hard engineering work sits in request shaping, valuation logic, state transitions, and the operator-facing layer that cleans up edge cases.
- Why keep it on the site
- Because it proves I can ship full-stack product systems, not only research-facing work.
- What matters technically
- Request shaping, valuation logic, state transitions, admin traceability, and the handoff between automation and human review.
Representative request shape
POST /api/trade-in/quote
{
"device_model": "iPhone 14 Pro",
"condition": "screen_minor_wear",
"storage_gb": 256,
"region": "CA-ON"
}
200 OK
{
"quote_id": "qt_1840",
"valuation_band": [410, 465],
"review_required": true,
"next_step": "checkout"
}Quote intake
A guided intake turns messy device details into a structured request before payment ever starts.
Valuation engine
Rules, condition handling, and review flags decide whether the result can be priced instantly or needs human review.
Order state
Checkout, payment, tracking, and after-order updates are treated as one stateful system rather than disconnected pages.
Admin review
The hard part is the operator-facing layer: exceptions, fulfillment, and traceability after the happy path ends.
Core toolkit
The stack is layered by how deep it runs, not by how nice it looks on a resume.
That is the visual difference between a generic full-stack profile and someone moving across research, software systems, and hardware context.
Core depth
Research / system substrate
- AI systems and model-to-interface thinking
- Sensing, telemetry, and monitoring pipelines
- Hardware-facing software and embedded-adjacent work
- Semiconductor reading track: silicon-based III-V and heterogeneous integration
Applied systems
ML, data, and operational software
- Python for ML and data-facing systems
- IoT dashboards, reporting, and operator surfaces
- SQL and decision-support data models
- MQTT, automation, and service integration
Delivery layer
Shipped product surfaces
- JavaScript and React interfaces
- PHP services and internal tooling
- Admin workflows, dashboards, and customer-facing flows
- LLM integrations when they solve a real interface problem
Field notes
I only publish notes when a system taught something worth keeping.
本地跑模型，我会先看 Gemma 4，也会拿 Qwen3.6-Plus 做参照
如果目标是本地部署，我先看的不是榜单，而是机器条件：显存多少、要不要离线、中文是不是主场景。这三个问题先定，选型就不会跑偏。Gemma 4 现在很值得看。Google 在 2026 年 4 月 2 日正式发布 Gemma 4，路线很清楚：开源、能本地跑，还把推理、函数调用、代码和多模态一起补强。对想做本地知识库、离线助...
AI 算法在网络安全里，真正有用的是这些地方
网络安全一直是高噪音行业。日志太多、告警太多、误报太多，真正危险的东西往往藏在最普通的一堆信息里。AI 算法放在这里，价值不是“自动接管安全”，而是先帮人把噪音压下去。我觉得现在最实用的几个方向很明确。第一是异常检测，尤其适合做账号行为、流量模式、终端活动的基线判断，帮助团队更快看见不合常理的变化。第二是钓鱼邮件和恶意...
Contact
For research, systems roles, or hard technical conversations, email works fastest.
A short note about the role, problem space, or system is enough.