About us
We put AI into our own product first, then we put it into yours.
Zhiyuan Qidian (Beijing) Intelligent Technology Co., Ltd. We did not come out of staffing. We built an AI product of our own first, learned the hard way what it takes to run AI inside a real business, and only then began doing that work for clients.
Who we are
We are a small company registered in Tongzhou District, Beijing, under unified social credit code 91110112MAKGTXAN85. We are not large, and we do not use size as an argument.
We run two lines of business. The first is our own product, kduanju.com — an AI short-drama production SaaS that integrates 10+ AI vendors and 40+ models across text, image, video and TTS, currently in invite-only beta. The second is FDE outsourcing: we embed forward deployed engineers at your site, or work remotely, to wire AI into the systems, permissions and workflows you already run, and hand over a production system you can accept.
Why we do this work
Building kduanju, we made the mistakes ourselves. The model choice gets reopened more than once, every vendor has its own reliability profile and billing quirks, and the output still has to pass through real review, publishing and permission workflows. Getting a demo to work and getting a system to hold up in production are two different jobs, and documentation does not teach the second one.
Our clients turned out to be stuck in the same place we had been. The model was never the bottleneck — nobody had wired it into the systems, permissions and workflows they actually run. That is work we know how to do, so we began doing it for clients.
We will not dress up the track record either: there is no illustrious history to point at, no client references we can publish yet, and our own product is still in beta. That is where we are today, written down here so you do not have to go looking for it.
Our position: what we will not do
Stating the limits is more useful than inflating the capabilities.
- We do not sell engineers by the day. Day-rate billing puts our incentive on making the project last longer, which is the opposite of yours.
- We do not take work whose real goal is cheaper hands. If you want to replace existing developers at a lower unit price, we are the wrong supplier, and we will say so.
- We do not take on projects whose acceptance criteria cannot be written down. "It works better" is not a criterion. "This step goes from X to Y" is.
- We do not exaggerate. No publishable references means no reference section; a product in beta is described as a product in beta.
How we decide whether a project is worth doing
The order never changes: four questions first, then any discussion of a solution. If one of them fails, we will tell you not to start yet.
- 01
1. Is this a standard requirement?
If an off-the-shelf product already covers it, buying that product is cheaper and faster than hiring us. In that case we will point you to the product.
- 02
2. Is there usable data?
Does the data exist, who holds it, and can you actually get to it? Without data, even a good design is just a demo.
- 03
3. Does someone own the outcome?
There has to be a business owner who can decide, and who is willing to let front-line staff shape the design. Projects pushed by IT alone, with the business sitting it out, go live and then sit unused.
- 04
4. What stays behind when we leave?
This is the line between FDE and ordinary outsourcing. The industry rule of thumb is blunt: if the first customer costs ten person-months and the tenth customer in the same sector still costs ten, the model has not worked. We hold ourselves to a second line as well: once a single client takes more than 30–40% of our deployment capacity, the business has quietly become high-end staff augmentation.
What a forward deployed engineer actually does
Neither of these figures is ours. The sources are given with each one.
47% / 31% / 22%
How forward deployed engineers spend their time: 47% with the customer, 31% writing code, 22% coordinating internally. 83% do not report into sales, and 0% carry a sales quota. The shorthand engineers in both China and the US repeat is "60% conversation, 40% engineering" (source: Perspective AI survey of 1,500 practitioners).
+21x
In the first half of 2026, FDE job postings on the Chinese professional network Maimai grew 21-fold year on year, and 37.3% of them came from AI-native companies (source: China Youth Daily, 2 September 2026). Average pay in China is RMB 408,000, with a median of RMB 366,000 (source: Liepin).
The team, and how much we take on
We are a small team and we take on only a few engagements at a time, deliberately. Once there are more projects than the team can personally stay close to, delivery degrades into sending a body to a desk — which is the thing we are trying not to be.
Company details you can verify
Every line below matches our business licence and permits.
- Registered name (Chinese)
- 北京智元启点智能科技有限公司
- Registered name (English)
- Zhiyuan Qidian (Beijing) Intelligent Technology Co., Ltd.
- Unified social credit code
- 91110112MAKGTXAN85
- Registered address
- Room 6328, Floor 3, Building 142, Zhuangchang Beili, Tongzhou District, Beijing, China
- ICP operating licence
- 京B2-20263361 (issued 2026-09-16, valid to 2031-09-16)
- Radio and TV programme production licence
- Held. Certificate details are available for verification on request.
- Our own product
- kduanju.com — Our own AI short-drama production SaaS, integrating 10+ AI vendors and 40+ models across text, image, video and TTS
- Website
- https://fde.kchangai.com
How to reach us
Email contact@mail.kchangai.com, add us on WeChat (the QR code is on the contact page), or fill in the form on the contact page. The first conversation is a 60-minute assessment of whether the work is worth doing. If the answer is no, we will say so plainly.
If any of this sounds right, the next step is one email
Tell us where the project is stuck, which systems it touches, and who owns the result. We will tell you whether it is worth doing — including when the answer is no.