Domain-specific knowledge

Not all procurements are created equal. A construction tender has completely different compliance rules than an IT services contract. The domain knowledge system teaches the AI the specific rules, terminology, and common pitfalls of each industry - so it catches issues that a generic analysis would miss.

Why it matters

Here's a real example. A construction RFP requires experience with 2 jaunbūve (new construction) projects. A vendor submits 2 projects, but one is actually pārbūve (renovation). Under Latvian construction law, these are legally distinct categories - a renovation does not count as new construction, no matter how large the scope.

A generic analysis would accept both as "construction projects" and mark the vendor as compliant. With construction domain knowledge loaded, the AI knows these are different legal categories and raises a Critical finding.

This is the kind of domain-specific nuance that matters in real procurement evaluations.

How it works

Domain detection

When an analysis starts, the system first determines the procurement's domain. This happens in two ways:

• Automatic detection (default) - the AI reads your RFP documents and identifies the domain based on CPV codes, legal references, and terminology. For example, references to "Būvniecības likums" (Construction Law) and CPV codes starting with 45xxx signal a construction procurement.
• Manual selection - when creating or editing a procurement, you can select the domain yourself from the dropdown. This skips auto-detection.

If the AI can't confidently identify a domain, the analysis continues normally without domain-specific rules. Nothing breaks - you just get the standard generic analysis.

What the AI learns

Once a domain is identified, the AI receives detailed knowledge about:

• Legal terminology - which terms have specific legal meanings and can't be used interchangeably
• Classification systems - how to categorize work types, building groups, experience categories
• Common mistakes - the most frequent compliance issues in that domain, with detection methods
• Validation rules - specific checks to perform, like cross-referencing experience claims against official certificates
• Regulatory references - which laws and regulations apply

This knowledge is injected into the main analyst, specialist agents, the verification agent, criteria extraction, and the compose agent — so every AI-powered feature benefits from domain-specific rules.

Qualification matching engine

Every domain analysis also includes a universal qualification matching engine. This cross-cutting module provides structured logic for validating vendor qualifications against RFP requirements:

• Matching modes - STRICT (exact match required), EQUIVALENCE (equivalent qualifications accepted), CATEGORY (category-level matching), and PREFERRED (soft preference, not mandatory)
• 3-source cross-validation - the AI verifies claims by checking the proposal text, supporting documents, and any available registry data before concluding
• Decision trees - standardized pass/fail logic that reduces inconsistency between analysis runs

This engine is automatically included with every domain prompt, ensuring consistent qualification assessment regardless of the industry.

Currently supported domains

Construction (Latvian context)

The construction domain covers Latvian public procurement for building and infrastructure projects. It includes:

• 7 legally distinct work types - jaunbūve, pārbūve, rekonstrukcija, restaurācija, atjaunošana, nojaukšana, konservācija - each with specific legal definitions that are not interchangeable
• Building group classification - 1st group (public/complex), 2nd group (residential), 3rd group (industrial)
• Experience validation - cross-referencing vendor claims against completion certificates (ekspluatācijas akts) and the state building registry (BIS)
• 13 common compliance mistakes - from work type confusion to certificate scope mismatches, with specific detection methods for each

Construction materials

Covers procurement of building materials and supplies with Latvian regulatory context:

• Material classification - structural, finishing, insulation, waterproofing, and specialty materials
• Standards compliance - EN and LVS standard references, CE marking requirements
• Common mistakes - specification mismatches, equivalence claims without proof, shelf life issues

Food and catering

For public procurement of food products and catering services:

• Hygiene and safety regulations - PVKN requirements, HACCP certification validation
• Classification - fresh produce, processed foods, prepared meals, catering services
• Common mistakes - missing food safety certifications, incorrect product categorization, shelf life and delivery requirement mismatches

IT and digital services

Covers information technology, software, and digital service procurements:

• Service classification - software development, system integration, IT infrastructure, consulting, managed services
• Qualification validation - industry certifications (ISO 27001, CMMI), personnel qualifications
• Common mistakes - vague scope definitions, missing SLA requirements, license compliance issues

Medical and pharmaceutical

For healthcare equipment and pharmaceutical procurements:

• Regulatory framework - medical device classifications, CE marking for medical devices, pharmaceutical distribution licenses
• Classification - medical devices by risk class, pharmaceuticals, laboratory equipment
• Common mistakes - missing conformity declarations, incorrect device classification, incomplete technical specifications

Technical equipment

Covers procurement of specialized technical and industrial equipment:

• Equipment classification - industrial machinery, measurement instruments, laboratory equipment, transport equipment
• Standards and certification - CE marking, calibration requirements, safety certifications
• Common mistakes - specification mismatches, missing warranty terms, incorrect technical parameters

Setting the domain

When you create or edit a procurement, you'll see a Procurement Domain dropdown:

• Auto-detect (AI) - the default. The AI will classify the domain when analysis starts.
• Construction, Construction materials, Food/catering, IT/digital, Medical/pharmaceutical, Technical equipment - manually select if you know your domain.

In most cases, leaving it on auto-detect works well. Manual selection is useful when you want to be sure, or if your RFP documents don't contain obvious domain signals.

The domain is set per-procurement, not per-bid. All bids within the same procurement are analyzed using the same domain knowledge.

Impact on results

When domain knowledge is active, you may notice:

• More precise findings - issues that reference specific legal categories rather than generic "missing information" notes
• Fewer false positives - the AI understands what terms mean in context, so it's less likely to flag something incorrectly
• Domain-specific recommendations - suggestions that reference the actual regulatory framework
• Provenance clarity - findings from domain-aware specialists are tagged with their source criterion or lot