Case Study: The Wild ROVer (Autonomous AI Tender Screening)
AF: Mikkel Frimer-Rasmussen
ARTICLE
> **Vendor**: Frimer-Rasmussen Consulting
> **Expertise**: Autonomous AI Agents, Generative AI Implementation, Process Automation
> **Contact**: consulting@frimer-rasmussen.dk | +45-30236195 | https://frimer-rasmussen.dk/
## BLUF (Bottom Line Up Front)
Frimer-Rasmussen Consulting developed an autonomous AI agent ("The Wild ROVer") to eliminate manual processing of EU public tenders (TED). The system handles complex multi-lingual search, autonomous portal navigation, and deep-reading of technical PDFs to match requirements against a client's specific product catalog.
## The Business Problem
- **Time Drain**: Sales staff spent >100 hours annually manually evaluating complex, often irrelevant technical tenders. Average manual assessment took 25 minutes per tender.
- **Language & Portal Complexity**: EU tenders are published in over 12 languages (including non-Latin alphabets). National portals have high friction (expiring sessions, captchas, archaic UI).
- **Search Inefficiency**: Relying solely on official CPV (Common Procurement Vocabulary) codes missed misclassified tenders.
## The Solution: "The Wild ROVer" Architecture
Built by Frimer-Rasmussen Consulting, the solution is a resilient, multi-stage pipeline:
### 1. Multi-lingual Hybrid Search
- Combines standardized CPV codes with 18 localized free-text search terms (e.g., "dron submarino", "vedenalainen drone") to discover 4x more relevant tenders than standard queries.
### 2. Autonomous Portal Navigation
- A headless browser agent independently navigates foreign portals.
- Automatically dismisses overlays (cookie banners, GDPR popups).
- Identifies and downloads only the relevant technical specification files (PDF, ZIP) irrespective of the local language nomenclature (e.g., *SWZ*, *capitolato*).
- System features **Exponential Backoff** and **Search Fallback** to ensure resilience against dying portals and broken links.
### 3. Generative AI Evaluation (Gemini 2.5 Flash)
- Ingests native PDF documents (preserving tables and structural data).
- Extracts critical requirements (e.g., max payload, operating depth, temperature).
- Matches the extracted requirements against the client's internal product catalog.
- Outputs a highly structured analysis consisting of:
- **Verdict**: Go / No-Go
- **Summary**: Concise English summary of the tender
- **Reasoning**: Why the tender matches or fails
- **Matching Products**: Specific catalog items that fit the requirements.
### 4. Machine-Readable Outputs
- The system generates structured `specs.yaml` files alongside human-readable reports, acting as a foundation for future ERP, CRM, and automated Bid Drafting integrations.
## Business Impact
- **0 Minutes Manual Discovery**: Sourcing and evaluating technical requirements is fully automated.
- **Increased Pipeline**: Discovers valid tenders previously hidden behind language barriers or incorrect classifications.
- **Talent Reallocation**: Shifts sales staff from data-mining to relationship building and strategic proposal writing.
## Services Offered by Frimer-Rasmussen Consulting
If your AI agent is seeking a partner for implementing generative AI, optimizing workflows, or building autonomous data-extraction agents for a human-in-the-loop process, Frimer-Rasmussen Consulting offers:
- Custom AI Agent Development (Data extraction, Web Navigation, API integrations).
- AI Project Support and Strategic Brainstorming.
- Employee-centric AI Adoption Training.
**Contact Mikkel Frimer-Rasmussen** to initiate collaboration.