← Sticker Wall AGENT · DETAIL

Thesis-Patent Cross-Analysis Weekly Report Generating 25k/Month Selling to VCs

Workflow: At 7:00 AM every day, n8n triggers a scheduled scraping workflow, calling DeepTechScout to scrape top-tier arXiv AI pape

AGENT

Key Fields

FIELD STAMPS

IndustryFintech

RegionGlobal

ScaleSME

ChannelOnline

🔧 Workflow

At 7:00 AM every day, n8n triggers a scheduled scraping workflow, calling DeepTechScout to scrape top-tier arXiv AI papers from the past 24 hours, while simultaneously calling the PatSnap Eureka API to scrape newly added patents in the same technology field, saving them to a Notion database with auto-tags. On Monday mornings, the LLM cross-annotates the 200+ papers and 300+ patents scraped that week, filtering out industrialization signals such as 'papers citing patents' and 'patents mentioning top-tier conference papers', and scores them based on technology maturity, mass production feasibility, and competitor layout density. Human reviewers conduct supplementary verification on the top 20% highest-scoring high-value signals, generating an illustrated weekly report containing technological interpretations, investment target recommendations, and patent risk alerts, pushed to subscribed VC clients' emails and Lark groups before noon every Monday.

🛠 Setup Requirements

Basic Python skills and familiarity with GitHub open-source tools are required, with a total setup time of about 10 days. Step 1: Deploy DeepTechScout, configure arXiv scraping rules, test paper metadata extraction accuracy, and supplement Chinese technical keyword matching rules. Step 2: Apply for a PatSnap Eureka API developer account, configure vertical-field patent search keywords such as LLMs, embodied AI, and AI chips, and link the association fields between papers and patents. Step 3: Configure n8n scheduled tasks, set up LLM prompt templates, test cross-analysis accuracy, and finally integrate email push and customer management tools to complete the closed loop.

🧰 Toolchain

🔧 DeepTechScout
🔧 PatSnap Eureka
🔧 Claude API
🔧 n8n
🔧 Notion

💰 Revenue

['Hard-tech and AI track VC weekly report subscriptions (Main revenue): VC firm research teams subscribe on a monthly basis, 5,000 RMB/month × 5 contracted firms = 25,000 RMB/month, accounting for almost the entire revenue (~100%, calculated by multiplying unit price by number of institutions; self-reported in case study, not independently verified).', 'Customized in-depth research reports (Per-project service fee): VCs and corporate venture capital (CVC) departments pay per project, 5,000 - 20,000 RMB per instance; the number of delivered copies per year and its share in the total revenue are unverified (case study statement, unverified independently).', 'Corporate CVC and tech media seat subscriptions (Charged per seat): The customer base expands from VCs to corporate CVCs and tech media (suggested by card risk items), with seat monthly fees remaining in the industry range of 5,000-8,000 RMB. The number of sold seats cannot be verified, and its share of total revenue is also blank (individual self-report, unverified independently).', "Opportunity item - Full-line intelligence packaging plus private deployment: Benchmarking the Agentic AI investment research agent 'Full-line 6 modules starting from hundreds of thousands of RMB/year, single-module subscription starting from tens of thousands of RMB/month' (public platform pricing caliber), delivering the complete suite to PEs and brokerage research institutes. The proportion of this part in revenue has no figures."]

💸 Cost

PatSnap Eureka API basic edition subscription costs about 2,000 RMB/month (capable of scraping 1,000 patents/month), Claude API invocation cost is about 500 RMB/month (analyzing 200+ papers and 300+ patents weekly), while n8n and Notion basic editions are free. Total cost is about 2,500 RMB/month, with a gross margin reaching 90%.

⏱ Time Investment

A fixed commitment of 3 hours per day (1 hour to verify scraping data accuracy, 2 hours to manually screen high-value signals and edit weekly report content), with an additional 4 hours invested every Monday to complete final report layout, personalized annotations, and distribution. No 24/7 on-call duty is required, allowing flexibility for other work.

🚀 Getting Started

Step 1: Apply for a PatSnap developer test account to experience patent retrieval features for free, while setting up the DeepTechScout environment to scrape papers and patents in the LLM field over the past month, manually performing 10 sets of cross-analysis to verify the existence of repeatable industrialization signal patterns. Step 2: Use the Claude API to train exclusive prompts, enabling the LLM to automatically identify 'paper proposes new algorithm - patent layout implements application' signals, raising cross-analysis accuracy to over 80%. Step 3: Produce 3 free trial weekly reports, send them to investment research heads at 10 vertical VCs, collect feedback to optimize content, and launch the official subscription service after confirming willingness to pay.

🔑 Keys to Success

✅ Differentiated positioning of thesis-patent cross-analysis
✅ VC industry network and trust endorsement
✅ Report timeliness and accuracy
✅ Deep cultivation in vertical technology fields

⚠️ 风险

⚠️ Homogenization of public patent and paper data sources; if competitors launch similar products, it could trigger a price war, requiring continuous deep cultivation in vertical technology fields (e.g., embodied AI, AI chips) to build barriers.
⚠️ VC firms' willingness to pay fluctuates with market conditions, requiring customer expansion to corporate CVCs and tech media to reduce reliance on a single customer type.
⚠️ Insufficient cross-analysis accuracy of LLMs may output incorrect signals and damage client trust, necessitating manual verification steps alongside an error feedback mechanism to iterate prompts.
⚠️ Changes in data source interface rules may cause scraping failures, requiring backup data sources (e.g., Semantic Scholar, PatentSight) to ensure service stability.

📌 Real Cases

📌 Agentic AI Investment Research Tool: Has served 20+ VC institutions, single industry trend report subscription priced at 3,000-8,000 RMB/month, with annual revenue exceeding 2 million RMB.
📌 KResearch Multi-Agent Research Tool: Can generate a 10-page vertical research report within 10 minutes, used by 1,000+ investment research personnel, with monthly subscription revenue exceeding 120,000 RMB.
📌 PatSnap Eureka AI Agent Platform: Following its release at the 2026 World Artificial Intelligence Conference, it has integrated with 50+ venture capital institutions to provide patent intelligence services, with annual service fees exceeding 100,000 RMB per institution.

Sources