Inspiration

Small and medium-sized businesses often miss valuable tenders, procurement notices, research collaboration opportunities, and government policy programs—not because the opportunities do not exist, but because the information is fragmented across thousands of university, research-institute, and government websites.

Manual monitoring is slow and expensive. Traditional aggregators create another problem: too much irrelevant information, repeated listings, and little understanding of what actually matters to a specific company.

We built 智慧商机信息管理系统 (Smart Opportunity Intelligence System) to turn scattered public information into personalized, decision-ready business opportunities.

What it does

The system continuously monitors user-selected websites and a reusable source library maintained by the platform administrator. It then:

• Collects tender, procurement, project, policy, subsidy, and research-cooperation information.
• Uses GPT-5.6 to extract key fields such as issuer, deadline, region, budget, qualification requirements, and contact information.
• Generates a concise summary, relevance explanation, and recommended next action for each opportunity.
• Matches opportunities to each company based on its industries, capabilities, keywords, and regions of interest.
• Deduplicates the same notice across multiple websites.
• Crawls a shared source only once, even when multiple companies subscribe to it, then distributes the result according to each tenant's rules.
• Lets users add a target through either a website URL or a natural-language instruction.
• Provides a summary dashboard and preserves the original source link for verification and detailed reading.
• Sends scheduled Feishu digests and can prioritize high-value or time-sensitive opportunities.
• Supports multi-tenant account management, source quotas, account validity periods, expiration reminders, and administrator-controlled permissions.

A pilot workflow is already monitoring selected Shaanxi science, technology, and industry-policy sources and delivering scheduled Feishu digests twice a week.

How we built it

We designed the product as a multi-tenant intelligence pipeline with five layers:

1. Source registry — stores canonical websites, crawl rules, update frequency, ownership, and tenant subscriptions.
2. Collection agents — OpenClaw-based tasks collect pages from configured websites and support user-defined URLs or text instructions.
3. AI processing pipeline — GPT-5.6 classifies notices, extracts structured fields, summarizes content, identifies deadlines and requirements, and generates company-specific relevance explanations.
4. Shared deduplication and distribution — normalized URLs, content fingerprints, and semantic comparison prevent repeated crawling and repeated alerts.
5. Application layer — users review summaries and source details, while administrators manage users, quotas, permissions, source reuse, and account expiration.

We used Codex with GPT-5.6 throughout the build to turn the initial business workflow into an executable product architecture, design the multi-tenant data model, implement and refactor collection and parsing logic, define structured AI outputs, generate tests, diagnose edge cases, and accelerate iteration across the backend and user experience.

Challenges we ran into

• Public-sector and university websites have inconsistent layouts, encodings, pagination patterns, and attachment formats.
• The same opportunity may appear on several websites with different titles or formatting.
• High recall is important, but sending too many irrelevant alerts quickly destroys user trust.
• AI summaries must remain traceable to the original notice and must not invent missing facts.
• Multi-tenant reuse must reduce infrastructure cost without leaking one company's private interests or configuration to another.
• Deadlines, account permissions, source quotas, and scheduled delivery all need reliable background-task management.

Accomplishments that we're proud of

• Built an end-to-end workflow from source monitoring to AI extraction, summarization, matching, and Feishu delivery.
• Designed a shared-source architecture that avoids duplicate crawling across tenants.
• Combined platform-managed information sources with user-defined targets.
• Added business-ready SaaS controls, including per-account quotas, validity periods, expiration reminders, and centralized administration.
• Validated the concept with a real policy-monitoring workflow instead of using only synthetic demo data.
• Kept every AI-generated insight connected to the original source so users can verify important details.

What we learned

The main value is not collecting more pages—it is reducing the time between discovering an opportunity and making a business decision.

We also learned that a scalable intelligence product should separate source collection from tenant subscriptions. This makes one crawl reusable across many customers while still allowing every company to receive a different relevance score, summary, and alert.

Finally, structured outputs and source-grounded summaries are essential. Users trust the system more when every deadline, qualification, and recommendation can be checked against the original notice.

What's next

• Expand from Shaanxi to nationwide universities, research institutes, military-industry research organizations, and government departments.
• Add stronger PDF, table, and attachment extraction.
• Build a company capability profile that automatically compares qualifications with tender requirements.
• Add opportunity scoring based on fit, deadline urgency, budget, competition, and historical success.
• Support instant alerts, daily digests, and weekly executive summaries across Feishu, email, and mobile channels.
• Add a feedback loop so user actions continuously improve ranking and matching.
• Provide organization-level analytics showing opportunity coverage, response speed, conversion, and missed-opportunity risk.

Built for

Track: Work & Productivity

Primary users: Small and medium-sized business owners, sales teams, business-development teams, consulting firms, and organizations that need to monitor public-sector opportunities efficiently.

Built With

• codex
• feishu
• gpt-5.6
• markdown
• multi-tenant-saas
• openclaw
• web-scraping