挑战
Finding reliable international suppliers was a time-consuming, manual process:
- Searching through thousands of supplier listings on B2B platforms was overwhelming
- No way to filter by social media presence or digital footprint quality
- Supplier information was scattered and required manual aggregation
- No AI-assisted guidance for sourcing decisions
- Bulk import/export of vendor data was unavailable
我们的解决方案
We built an AI-powered vendor discovery platform with intelligent search, automated data enrichment, and an AI chat assistant for sourcing guidance.
Architecture
- Backend: Node.js/Express with TypeORM and PostgreSQL
- Search Engine: Elasticsearch for fast, full-text vendor search with priority filtering
- Frontend: React 18 + Vite with Redux Toolkit and Material UI
- Admin Dashboard: Dedicated admin interface for vendor management
- AI Assistant: OpenAI + Google Gemini-powered chat for sourcing guidance
- Scraping Engine: Puppeteer + Selenium for automated vendor data collection
Key Features
- Intelligent Search - Elasticsearch with filters for name, category, location, social media
- AI Chat Assistant - Conversational AI for vendor discovery and sourcing advice
- Social Media Extraction - Automatic indexing of Instagram, Facebook, LinkedIn, Twitter, YouTube, TikTok, Pinterest profiles
- Data Enrichment - Auto-extraction of contact info, brands, certifications, MOQ
- Bulk Operations - CSV import/export for vendor data management
- Vendor Self-Onboarding - Suppliers can register and manage their own profiles
- Favorites & History - Save preferred vendors and track search history
- Subscription Billing - Stripe-powered credit-based payment system
Data Pipeline
- Scraping - Puppeteer/Selenium collects vendor data from B2B platforms
- Enrichment - AI extracts structured data (contacts, certifications, social links)
- Indexing - Elasticsearch indexes enriched data with priority-based ranking
- Search - Full-text search with category, location, and social media filters
- AI Guidance - Chat assistant helps users refine searches and evaluate vendors
成果
技术栈
常见问题
MicrocosmWorks built a multi-signal ranking engine that combines NLP analysis of vendor capabilities, historical performance data, certification verification, and geographic proximity into a weighted composite score. The AI model was fine-tuned on procurement outcome data to learn which vendor attributes correlate most strongly with successful engagements.
Yes, MicrocosmWorks implemented a web crawling pipeline that continuously indexes vendor websites, trade publications, and industry directories beyond standard databases like ThomasNet. The NLP classifier can identify vendor capabilities from unstructured text, discovering suppliers that do not appear in curated databases but match the buyer's technical specifications.
MicrocosmWorks designed an automated re-crawling schedule that revisits vendor profiles every 30 days, with change detection algorithms that flag significant updates like new certifications, facility expansions, or leadership changes. Stale profiles are deprioritized in rankings, and buyers receive alerts when a previously recommended vendor's profile changes materially.
MicrocosmWorks built REST API connectors for SAP Ariba, Coupa, and Oracle Procurement Cloud, allowing discovered vendors to be pushed directly into the buyer's existing sourcing workflow. The platform also exports vendor shortlists in standard formats compatible with any ERP system's supplier master data import process.
MicrocosmWorks builds AI-powered procurement platforms at rates of $25-$50/hr, with a full vendor discovery system including the crawling pipeline, NLP ranking engine, and ERP integrations typically requiring 600-900 development hours. The AI model training and fine-tuning phase usually accounts for 100-150 hours of that total.