FintechGlobal
← Sticker Wall
AGENT · DETAIL
Multi-agent paper scraping for cutting-edge insights sold to VCs, generating 20k RMB monthly
Workflow: The multi-agent workflow is automatically triggered every midnight, first scraping cutting-edge papers and public patent
AGENT
Key Fields
FIELD STAMPSIndustryFintech
RegionGlobal
ScaleSME
ChannelOnline
🔧 Workflow
The multi-agent workflow is automatically triggered every midnight, first scraping cutting-edge papers and public patent data from the past 24 hours from academic paper repositories like arXiv and patent databases (e.g., Patsnap). Large language models then extract technological breakthroughs, potential commercialization scenarios, and competitor dynamics. Human operators need to fact-check the key conclusions output by the model—especially judgments on technological maturity and the commercial feasibility of application scenarios—filtering out false correlations and overly optimistic predictions. Finally, a structured cutting-edge technology briefing is output and sent weekly to contracted early-stage investment institution investment managers, while synchronizing to follow up on their customized research demands.
🛠 Setup Requirements
First, you need to master basic multi-agent framework deployment capabilities and be able to debug rules and filtering conditions for data scraping modules. There is no need to master underlying code writing; knowing how to call open-source components is sufficient. In the early setup stage, you need to select 1-2 vertical niche fields (e.g., embodied AI, AI drug discovery) to run through the entire process of data scraping, report generation, and manual verification, which is expected to take 2-3 weeks to complete the first usable version of the workflow. Afterwards, you only need to regularly update data source rules and optimize prompt templates, resulting in extremely low maintenance costs.
🧰 Toolchain
- 🔧 Lobster-Research open-source multi-agent research framework
- 🔧 KResearch multi-agent research tool
- 🔧 arXiv open academic interface
- 🔧 Claude large language model
- 🔧 Patsnap patent database interface
💰 Revenue
① Early-stage VC institutions subscribe to cutting-edge technology briefings on a monthly basis (main revenue): Institutions pay per article/monthly, 5,000 RMB per article × 4 deliveries per month = 20,000 RMB basic monthly subscription. Signing 3-5 institutions accounts for about 40%-100% of monthly revenue (derived from internal numbers, case study caliber, without independent verification); ② Customized deep research projects for niche tracks: Institutions pay an additional 10,000-30,000 RMB per project. The actual number of closed projects has no verifiable figures, and the proportion of total revenue is not specified; ③ Industry map and data source module subscription value-added: Peer product single-module subscriptions start at tens of thousands of RMB per month; industry map construction is shortened from 2 weeks to 1 day (case source, without independent verification), with additional subscription fees per module. The number of purchased modules cannot be verified, and the proportion has no public caliber; ④ Opportunistic revenue—Multi-agent research report generation framework licensing: Licensing Lobster-Research-type frameworks to small and medium-sized investment research teams per seat. Seat prices have not been publicly disclosed, and there is currently no data on how much revenue it accounts for.
💸 Cost
The main cost is the API fees for calling large language models, which is about 500-800 RMB per month; if using a paid patent database interface, an additional 200-300 RMB in monthly costs is added. There are no other fixed expenses such as venue or labor, and the marginal cost is close to zero.
⏱ Time Investment
Invest 1.5-2 hours daily. In the morning, monitor whether the automatically scraped data stream has anomalies; in the afternoon, spend 1 hour on manual verification of core conclusions and report typesetting; take 2 hours each week to connect with investment institutions' customized demands.
🚀 Getting Started
Step 1: Clone the code repository of open-source multi-agent research frameworks such as Lobster-Research, deploy locally and familiarize yourself with basic operations, select a technology niche you are familiar with (e.g., semiconductors, AI applications), and configure the scraping rules for arXiv and patent databases. Step 2: Use the framework to run through 3-5 test reports, focusing on polishing prompts to ensure the output technology commercialization analysis matches the reading habits of investment institutions while eliminating redundant academic expressions. Step 3: Compile the test reports into a sample booklet and send it via LinkedIn and venture capital communities to 20 venture capitalists focusing on early-stage tech investment, providing 1 free trial report in exchange for cooperation intent.
🔑 Keys to Success
- ✅ Depth of human judgment on technology commercialization pathways in adjudication
- ✅ Completeness of scraping multimodal cutting-edge data sources
- ✅ Building pay-per-use trust relationships with investment institutions
- ✅ Vertical field focus in cutting-edge technology screening
⚠️ 风险
- ⚠️ Heavy reliance on open-source frameworks may introduce system stability risks; if the framework stops updating or interfaces change, the workflow will be interrupted
- ⚠️ If investment institutions build similar internal research tools themselves, it will directly terminate outsourcing demand
- ⚠️ If report accuracy suffers misjudgment, especially regarding technology commercialization pathways, it may damage long-term cooperative credibility with investment institutions
- ⚠️ Improper handling of copyright boundaries for cutting-edge patents and academic papers may trigger data compliance risks
📌 Real Cases
- 📌 A Silicon Valley independent entrepreneur reported by 36Kr operated solo relying on a multi-agent investment research framework, achieving annual revenue of 72 million RMB and successfully raising 200 million RMB in financing, verifying the commercialization potential of personal investment research services
- 📌 Aicue AI Research Agent has served nearly 100 domestic early-stage VC institutions, stably delivering over 200 cutting-edge technology insight briefings per month, with single-client annual payments reaching 30,000-100,000 RMB
- 📌 Patsnap Eureka AI Agent platform was released at the 2026 World Artificial Intelligence Conference, having helped multiple investment institutions shorten patent research cycles from 7 days to 4 hours, with a paid conversion rate of 35%