Labs - Research

RAG Professional

AI Knowledge Retrieval Systems for Professional Domains

In professional fields like patents, law, medicine, and finance,
we research RAG systems that are more accurate and reliable than web search

View Basic RAG Technology View Domain Applications

Why Domain-Specific RAG?

Patent filing, legal consulting, medical research, financial analysis...
To leverage AI in professional domains, simple web search has its limits.

By building professional data into a hybrid knowledge base (Vector+RDB+Search),
you can leverage more and more accurate information reliably than web search.

Web Search vs Hybrid RAG

Category	Web Search	Hybrid RAG
Reliability	Unclear sources, hallucination	File (source) based evidence
Accuracy	Simple keyword matching limits	Vector (semantic) + RDB (filter) + Search (lexical)
Freshness	Crawling time lag	Real-time RDB/Memory updates
Security	Data exposure risk	Local/dedicated server + access control
Context	Single-turn processing	Memory-based conversation context

Domain-Specific RAG Applications

Building domain-specialized data into vector databases

Patent

In Use

Patent law, examination guidelines, case law, prior art

Patent Act / Enforcement Decree
Technology-specific examination guidelines
Patent court rulings
KIPRIS registered patents

View Service

Legal

Research

Statutes, case law, contracts, legal consulting

Laws / Enforcement Decrees
Supreme/lower court rulings
Standard contract templates
Legal interpretation cases

Medical

Research

Research papers, clinical trials, guidelines

Medical papers/reviews
Clinical trial data
Practice guidelines
FDA/MFDS approval information

Finance

Research

Financial regulations, supervisory rules, interpretations

Financial regulations
FSS supervisory rules
Authoritative interpretation cases
Financial product terms

Patent RAG Hybrid Layers

5-layer hybrid storage structure for patent domain

File

Source Storage

Patent specification PDF/XML
High-resolution drawings
Evidence document originals

RDB

Metadata (Filter)

Application number, date, applicant
IPC/CPC classification codes
Legal status (registered/rejected)

Vector

Semantic Search

Claim technical essence embeddings
Problem-solution semantic vectors
Similar patent clustering

Keyword Search (Lexical)

Patent full-text index
Proper noun/abbreviation dictionary
BM25 ranking algorithm

View Patent Service

MCP-Based Service Deployment

Delivering RAG systems as AI-native APIs

Local MCP

stdio communication, installable. Suitable for sensitive data

Personal/internal use

Remote MCP (SSE)

HTTP-based, SaaS-ready. API key/OAuth authentication

External service delivery

Hybrid

Sensitive data local, common data remote

Enterprise customers

MCP RAG Server Architecture

Clients

Claude Code Claude.ai Custom App

MCP Protocol (JSON-RPC)

MCP RAG Server

search_patent() find_similar_claims() get_guidelines()

Hybrid Storage Layer

Patent Law DB Examination Guidelines DB Case Law DB Prior Art DB

Technical Considerations

Key factors to consider when building domain-specific RAG

5-Layer Hybrid Storage

Organic integration of File (source), RDB (meta), Vector (semantic), Search (lexical), and Memory (context).

RDB Metadata Filtering

Pre-filtering with structured data like permissions, dates, and authors dramatically improves search speed and accuracy.

Keyword + Vector Hybrid

Keywords for proper nouns (model names, error codes) and vectors for context and intent — complementary search.

Memory & Context

Maintaining previous conversation context in Memory for seamless answers to follow-up questions.

Want to collaborate on research?

We welcome collaboration on building
domain-specific RAG systems and research partnerships