Building an Investment Scoring Page with Qwen

I spent two days trying to use Qwen3.8-Max-0902 to build a functional investment project scoring page from scratch. Not writing a paper, just making a simple page: input a few scores, automatically calculate the total. When I was doing product work at big tech companies before, scheduling for this kind of page often took two weeks. Now the model can scaffold it out in one go. The model is basically an AI program that generates text or code based on prompts.

First, let's talk about the problem. Early-stage projects look at people and direction, but every time after a chat, you have to record a bunch of subjective scores. For a small project I invested in under my name, I know the founder personally. Previously, I used spreadsheets to fill things in manually, and the data was scattered. The frontend page is the part users see and click in their browser. It usually consists of three things: HTML is the skeleton, CSS is the skin, and JavaScript is the action behind the buttons. Putting all three into one file allows you to open it with a double-click, which is great for a minimum viable demo version.

This upgrade to Qwen3.8-Max-0902 involved extra training around programming and professional office tasks, as mentioned officially. It also notes suitability for complex enterprise tasks, scientific research, and long-cycle tasks. Its context length reaches 1M Tokens. Tokens can be roughly understood as the amount of text the model can read/write at once; longer contexts make it easier to remember previous requirements. Building a page doesn't require stuffing in a lot of material at once, but it helps when there are many rules.

The plan is to let the model generate a single-file scoring page. Single-file means all code is in one file. Beginners should follow the steps below.

1. Open the Qianwen web client, log in, and click New Conversation. An empty input box usually appears in the interface. Expect to see a conversation area with no old content.

2. Find Qwen3.8-Max-0902 in the model selection area. If the entry only says Qwen3.8-Max, select that first, then ask in the conversation to handle it using the programming capabilities of the 0902 version. Expect the current model name to appear below the input box.

3. Paste the prompt and click Send. CDN in the prompt refers to style files hosted on someone else's server, which might not be accessible locally. Here is the prompt:

Help me write a single-file webpage for early-stage project scoring. Fields include: Founder Background, Market Size, Product Completeness, Data Validation, Team Execution. Weights are 25, 25, 20, 20, 10 respectively. Each field accepts inputs from 0 to 10, automatically calculates the weighted total score, and displays a rating: A for above 80, B for 60-79, C for below 60. Requirements: Use only native HTML, CSS, JavaScript. Do not reference external CDNs, do not rely on the network, do not split files, do not explain, output complete code only.

4. Wait for the response. I waited about one to two minutes, but stability isn't guaranteed. Expect to see a block of code wrapped in PFF0html and the trailing ` , otherwise the browser will display it as plain text and won't execute the code. Another pitfall is that if you type text into the input boxes, the score calculation shows NaN, meaning it can't compute a number. Beginners can add a line to the prompt: "Input boxes allow only numbers and perform boundary checks," which restricts input to 0-10.

In terms of results, this page is very plain, with no login and no database. But it demonstrates the scoring logic. From an investment perspective, the page itself has no moat. The model lowers the barrier for frontend development; what's truly valuable is how scoring data accumulates, how projects are followed up, and how relationships with founders are maintained. Qwen3.8-Max-0902 topped the frontend programming leaderboard this time; the commercial significance is lower delivery costs. If early teams only know how to wrap a chat box, valuation is hard to justify; if they can integrate models into project due diligence, decision-making, and post-investment management processes, forming a private data loop—meaning data and processes only this company can accumulate—then there is retention. Team execution is key; turning a small page into a usable workflow for clients within two days is more important than painting grand visions.

Next Steps could try two directions. One is putting the page on static hosting, meaning making the file accessible online for others; two is asking the model to add an "Import CSV" button. CSV is a comma-separated table file that can import a batch of project records. Don't rush to connect APIs yet; APIs are interfaces for programs to call each other. Get one file running smoothly first.