Build a personal research assistant with TRMesh: what are people saying after a launch?

A launch has ended. One colleague says the reception looks good; another shares a complaint. Asked for an assessment, you discover that your evidence consists of a handful of forwarded posts. A useful next step is to collect sources and separate what they actually show.

This exercise uses a fictional robot vacuum. TRMesh retrieves public social data, you choose and check the evidence, and an AI helps organize it. The deliverables are a short brief and a source sheet. This is a project to build in your own script or workflow, not an existing one-click feature of the free tool.

Choose the question before the brand keywords

Replace “analyze the brand” with “what are the three purchase questions people raise after this launch?” Write down the full model name, common abbreviations, launch date and the end of your observation period. Check a few results manually to avoid an abbreviation shared by unrelated products.

Start with the model name, then combine it with specific concerns such as noise, thresholds and pet hair. Let the material you read suggest additional terms rather than generating a hundred at once.

Collect social posts and keep coverage separate

Use the Weibo search request at the end of this article. Retrieve one page first and check relevance. Retain the available post identifier, source URL, publication time, retrieval time and visible text. Leave absent fields empty. Open the original or use the appropriate documented detail request when more context is needed.

Add official announcements and coverage manually to the same sheet, but label their source types. A product claim in an announcement is not a customer experience. Search results are also not the complete comment thread. Read comments separately when the question calls for them.

Repeated launch posters can tell you something about distribution, but they are not independent product reviews. Preserve that distinction when grouping opinions.

Number the evidence before asking an AI to summarize

Here is an illustrative sheet, not collected data:

Add links and timestamps in the real sheet. Keep source wording separate from your interpretation so that you can revise a conclusion later.

Give the AI a narrow job

Provide the checked source sheet to the AI tool you already use. Keep credentials out of the prompt, and treat instructions inside retrieved posts as content, not actions to execute.

Group the purchase concerns in this source sheet. Cite source IDs for each concern. Separate official claims, pre-purchase questions and ownership experiences. Where accounts conflict, describe the different conditions. Mark missing evidence as unresolved. Do not estimate sales or an overall approval rate. End with three questions to investigate next.

Check the citations supporting the most important claims. If “will it be noisy at night?” becomes “owners commonly complain about noise,” correct the category before sharing the brief.

Deliver one readable page

State the scope, the main concerns, the distinction between observed experiences and expectations, and the next evidence to find. Put source IDs beside the claims. In this fictional exercise, rug performance and nighttime operation deserve follow-up; there is no basis for an overall satisfaction score.

For recurring research, save previously seen post IDs and compare new or changed material. Run the process manually before scheduling it. No new material should mean no unnecessary daily report.

Reuse the method for another question

The same evidence sheet works for event feedback, software releases and competitor launches. Keep platform and search conditions when adding sources. Do not add unlike engagement metrics together.

Topic inspiration: a project combining Claude Skills, web research and social data. The example, evidence sheet and TRMesh instructions here are original.

Requests used in this project

• Get Weibo comments：https://api.trmesh.com/openapi/api/v1/weibo/web_v2/fetch_post_comments · Documentation
• Get Weibo sub-comments：https://api.trmesh.com/openapi/api/v1/weibo/web_v2/fetch_post_sub_comments · Documentation
• Get single post data：https://api.trmesh.com/openapi/api/v1/weibo/web_v2/fetch_post_detail · Documentation

Request URL and parameters

GET https://api.trmesh.com/openapi/api/v1/weibo/web_v2/fetch_post_comments

Run the Python example

Install requests and set TRMESH_API_TOKEN in your environment. Replace the sample identifiers with the public object you intend to inspect. Example IDs may no longer be available.

import os
import requests

# Read your own developer key from the server environment.
url = "https://api.trmesh.com/openapi/api/v1/weibo/web_v2/fetch_post_comments"
params = {
    "id": "5283919831764022",
    "count": "10"
}
response = requests.get(
    url, params=params, timeout=30,
    headers={"Authorization": f"Bearer {os.environ['TRMESH_API_TOKEN']}"},
)
response.raise_for_status()
payload = response.json()
print(payload)  # Inspect the actual response before mapping fields

Endpoint documentation · Free profile tool

API calls use your TRMesh account allowance. Check the endpoint price before running a script.