Resumes sent out disappear into thin air, staring at a blank document while writing reports makes your head hurt, and getting a sales table full of jumbled data leaves you clueless—these are the daily stress moments for professionals. Yesterday, these situations were deconstructed in front of the audience by Qwen APP in Wuhan.

This AI job-hunting practical class was guided by the Wuhan Human Resources and Social Security Bureau, hosted by Qwen APP and Wuhan Release. Instead of sticking to dry theories, it took the on-site job seekers and fresh graduates through three tasks: resume diagnosis, business report writing, and sales table analysis, using AI step by step. Turning materials into files, rather than just adding more text, is how Jin Shixing, the product manager of Qwen APP, assessed the value of AI documents, setting the tone for the entire course.

image.png

When it comes to writing a resume, Qwen suggests clarifying five things: provide all materials, explain the goal, define standards, set boundaries, and finally request an editable file. Many people just say "help me write a paragraph," but to create a good resume, you need to treat scattered experiences as raw materials. Use the requirements of the target position as a benchmark, and reinforce relevant facts in reverse. Turn a plain description like "participated in organizing an event" into a more powerful expression with background, actions, and results, which immediately increases persuasiveness.

Qwen also presented a set of three-step prompt words that can be directly replicated as a template. The first step is only for diagnosis. Let the AI read the attachment first, without rushing to rewrite it, but instead summarize the job requirements into no more than seven core requirements sorted by importance. Then establish a five-column comparison table: job core requirements, evidence already in the resume, evidence strength, gaps or unconfirmed information, and suggested actions. Any information not in the attachment should be marked as "not provided," strictly avoiding speculation or fabrication, and finally give three most urgent revision suggestions.

image.png

The second step is to rewrite based on factual information. Only use real content from the original resume, rewriting personal summaries and work experience into action, methods, and results structures. Keep each section to three to four points, and keep each point to no more than 65 Chinese characters. Mark any evidence not in the material as "to be supplemented" and include a对照表 (comparison table) showing the modified content and its factual sources for verification by the user. The third step is to generate a formatted Word resume. There are clear rules for file names, margins, font size, line spacing, and module title styles. After generation, it requires self-checking for page breaks, fonts, and punctuation.

Being suddenly handed a pile of unfamiliar files and being asked to make a PowerPoint presentation at short notice is another major nightmare in the workplace. Qwen APP's product manager Jingjing demonstrated a three-step emergency solution. The first step is to complete the context of the task, clearly explaining four types of background: the objective and theme, the presenter and audience, supporting materials, and the presentation format. For example, when reporting on a short drama transformation research to the team leader, it is necessary to clarify the core conclusion that short dramas are a key breakthrough direction, the role of the planner, the focus of the report on why they are suitable for the self-media transformation, how to quickly gain traction, and the monetization path, as well as the form of 15 pages for a 15-minute presentation. The second step is to let the AI read all the files and summarize them, then restructure the content according to the speaker's department responsibilities and recent business priorities, generating a draft. The third step is to use Qwen's four skill kits for fine-tuning, making the PPT more personalized.

The most challenging part is using AI to extract key metrics from messy data. Jin Shixing demonstrated a fictional case of a tea shop's operations, walking through the entire process of building, organizing, calculating, analyzing, and presenting. When given a raw sales detail sheet, first clean up the messy data while keeping the original table, unify the naming format, then input formulas to calculate gross profit and gross margin. This gives a foundation for business review. Finally, a sales table with 486 rows of messy data is condensed into a single clear PPT page with clear conclusions.

The accompanying six-step data analysis prompt words break down this methodology in great detail. The first step is to establish an analysis workspace, processing only the uploaded files without supplementing online information or speculating on missing facts, retaining the original data, and ensuring all conclusions can trace back to the data and calculation methods in the table. Create five workbooks: cleaned data, data processing log, indicator summary, business report, and definition explanation, and set filters, freeze headers, and standardize formats.

The second step is to clean the original data, unify the aliases of three stores and four channels, standardize product names, convert dates to YYYY-MM-DD, transform amounts with currency symbols or text formats into numbers, delete completely duplicate records, mark missing items as "to be confirmed" instead of guessing, and output the number of rows before and after processing, the number of various issues, and pending confirmations.

The third step is to enter traceable formulas, using Excel formulas rather than static results to fill in standard sales, product cost, gross profit, gross margin, discount rate, average revenue per cup, and date type indicators. Handle division by zero and missing values with division, retain two decimal places for amounts and one decimal place for ratios, and simultaneously record the indicator definitions in the definition explanation and generate a general verification table.

The fourth step is to conduct business analysis using natural language, summarizing by store, product, channel, and time period, answering questions such as which store has the largest scale, whether the highest sales also mean the highest gross profit, which channel has large volume but low gross profit, and which two periods contribute the most revenue. Each conclusion lists the data, the comparison object, and the calculation basis, and reminds not to generalize hypothetical samples into long-term trends.

The fifth step is to generate a one-page business report, displaying key indicators at the top, producing four charts: store revenue comparison, product cup count and gross profit comparison, channel gross profit comparison, and time period revenue comparison. Each chart title directly expresses the main finding instead of using vague titles. Below each chart, list three data insights and three action recommendations. The sixth step is for the data auditor to do the final check, verifying the number of rows before and after cleaning, whether all summaries equal the total, whether formulas cover everything, whether percentage denominators are consistent, whether charts are misleading, and whether conclusions exceed the data range. Output a four-column list of checking items, results, evidence, and matters requiring manual confirmation without directly modifying the data.