01Supply Chain Tracing
Query
“Trace the supply chain for a specific part 3 levels deep”
Answer
12 partners · 27 contracts · 3 risk nodes identified
Why · Single-source dependency · discontinuation risk · alternative suppliers surfaced instantly
ONTOFLOW · Ontology Platform
Korean-built Foundry,
Korean-language Ontology.
Structure scattered industry knowledge into an ontology-based Knowledge Graph, query it in Korean natural language, and validate every answer with the executed Cypher. Delivers equivalent value to Palantir Foundry at a fraction of the cost.
Ontology · Knowledge Graph · GraphRAG
Use Cases
Questions like these become answerable.
Ask in natural language, get instant answers, verify with the executed query. Safe even in audit and regulatory environments.
02Relationship Exploration
Query
“Find all partners shared by two procurement clients”
Answer
17 total — direct 7 · 1st-degree 6 · 2nd-degree 4
Why · Cross-order patterns · negotiation leverage · market concentration analysis
03Technology Gap Analysis
Query
“Which companies above TRL 7 hold a specific technology?”
Answer
24 qualifying companies · average TRL 8.2
Why · Instantly identify commercializable companies · automate technology matching
04Anomaly Detection
Query
“Top 5 technology fields with new entrants in the past 6 months”
Answer
1st: Quantum sensors (12 new) · 2nd: Digital twin (8) ...
Why · Market trends · early detection of emerging threats and opportunities
05Knowledge Expansion
Query
“Absorb this report (URL) into the graph”
Answer
27 entities extracted · 41 candidate relationships pending human review
Why · Auto-ingests reports, papers, and news · no gaps from missed updates
06Transparency Verification
Query
“Show me the evidence behind this answer”
Answer
Executed Cypher + 8 source nodes + confidence 0.94
Why · Audit and regulatory compliance · 0% LLM hallucination
Six Core Capabilities
Query · Expand · Visualize · Analyze · Verify · Synthesize.
01
Natural Language Search
Natural Language Search
Automatically converts Korean questions into graph queries (Cypher). A pattern-matching-first architecture handles 99.7% of questions without any LLM call.
100%
Routing accuracy (100%)
0.3%
LLM fallback rate
1.2s
Average response time
02
Knowledge Expansion Pipeline
Knowledge Expansion
Automatically extracts entities from external documents (reports, papers, web pages) and matches them against the existing Knowledge Graph to surface new relationships and insights.
5-stage
Automated pipeline
HITL
Human-in-the-Loop validation
Sandbox
Safe preview merge
03
Relationship Visualization
Relationship Visualization
Visualize an entire domain ecosystem on one screen with a 4-tier radial hierarchy and force-directed graph. Click any node for interactive 2-hop and 3-hop exploration.
4-tier
Radial hierarchy
Interactive
Re-query exploration
Live
Real-time graph
04
Geospatial Map
Geospatial Map
View real-location clusters of domain entities on an interactive MapLibre GL map. Highlight filtered and natural language search results directly on the map.
Filtering
Sector · cohort · grade
Clustering
Geographic concentration analysis
MapLibre
Open-source based
05
Workshop Dashboard
Realtime Analytics
Real-time Neo4j Cypher-query-based dashboard. KPI cards, distribution charts, collaboration rankings, and growth trends — all on one screen. No mock data — all live measurements.
Cypher
Real-time query based
0 mock
All live data
PLG
Prometheus · Loki · Grafana
06
Synthetic Data
Domain-aware Synthetic Generation
Generates validation, simulation, and training synthetic data grounded in domain ontologies. Delivers the same data quality even when real data is scarce or sensitive. TTA synthetic data standard compatible.
Validation
Algorithm regression testing
Simulation
Virtual scenario data
Privacy-safe
Synthetic replacement for sensitive data
System Architecture
From client to data and AI layers.
Clean separation across four layers: Frontend, API, Data, and AI. Every layer is open-source based and can be transitioned to self-operated infrastructure.
5-Stage Knowledge Expansion Pipeline
External documents become graph nodes automatically.
Input a URL, PDF, or HWP file and the 5-stage pipeline handles the rest. The final stage requires human validation (HITL) before merging.
Palantir Foundry vs OntoCraft
Same paradigm, greater depth.
We re-implemented Foundry's proven paradigm with Korean domain depth. The value Palantir delivers to the US government — at Korean SME pricing.
Category
Palantir Foundry
OntoCraft
Category
Data Sovereignty
Palantir
US cloud
Subject to ITAR/EAR
OntoCraft
On-premises in Korea
Fully self-sovereign
Category
Korean Language Support
Palantir
English-centric
Limited Korean
OntoCraft
Korean NL 100%
100% accuracy
Category
Deployment Time
Palantir
6–12 months
FDE consulting required
OntoCraft
2–4 weeks
Self-configurable
Category
Vendor Lock-in
Palantir
High
Proprietary format
OntoCraft
None
Open-source, self-operated
Category
Domain Specialization
Palantir
General (optimized for US military)
Global standard
OntoCraft
Korean industry-specific
6-domain dictionary
Category
Cost (3-year TCO)
Palantir
Baseline
License + FDE + infrastructure
OntoCraft
Far lower
Lower TCO for equivalent value
※ Palantir figures are self-calculated estimates based on public Foundry Enterprise disclosures and domestic cases — for reference comparison only.
Domain Coverage
Six domains, 462 Korean terms.
We build a dedicated dictionary for each domain — going deep, not broad, when entering a field.
82
Defense
Korean terms
100
Healthcare
Korean terms
90
Finance
Korean terms
60
Public Sector
Korean terms
70
Manufacturing
Korean terms
60
Education
Korean terms
462 terms total · ISO standard compatible · additional domains in progress
Technology Stack
Enterprise-grade open-source foundation.
On-premises deployable · air-gapped environment compatible · zero vendor lock-in.
FrontendReact 18 · TypeScript · Vite · MUI
BackendNestJS · TypeScript strict · 84 modules
Graph DBNeo4j · Cypher · industry standard
RDBPostgreSQL · audit history · automation rules
CacheRedis · 30/60s TTL · rate limiting
AI / LLMClaude · GPT · Ollama multi-model
MLFastAPI · Sentence Transformers
Vizvis-network · MapLibre GL · Recharts
InfraDocker · Kubernetes · Helm
TS DBInfluxDB · Telegraf · time-series processing
ObservabilityPrometheus · Grafana · Loki · Kubernetes monitoring
Quality Metrics
Measurable numbers, not marketing copy.
9,806+
Unit tests
232+ suites
100%
NL routing
All passed
strict
TypeScript
noEmitOnError
0
any allowed
ESLint enforced
93
Security tests
OWASP Top 10
163
E2E checks
Bash + Playwright
JWT
+ RBAC
6 roles · 65 permissions
Docker
One-click deploy
Compose · K8s
Frequently Asked Questions
Things you may be wondering.
Can it be deployed on-premises?+
Yes. The full stack comes up with a single Docker Compose command. Neo4j, PostgreSQL, Redis, and Ollama all run on-premises. Works identically in air-gapped environments.
How much does it use Claude/GPT APIs?+
Pattern matching handles 99.7% of queries, so LLM API calls occur at just 0.3%. In air-gapped environments, local LLMs such as Ollama can replace them 100%.
Does it integrate with existing systems (C4I, ERP, CRM)?+
Yes — REST API, GraphQL, WebSocket, and MCP standards are all exposed. A Palantir Foundry-style connector architecture supports simultaneous integration of 10+ data sources.
What if our Korean domain vocabulary is insufficient?+
We currently maintain dictionaries for 6 domains (462 Korean terms). New domains are built collaboratively during the PoC phase — domain expansion is a core area of collaboration.
What security level does it meet?+
JWT + Keycloak SSO · MFA · 6-role × 65-permission RBAC · full API audit logging · Cypher injection defense · ReDoS prevention. SOC2/ISMS certification planned for 2028.
What is the general pricing?+
Pricing is negotiated per project. SME: setup ~100M KRW + annual license 50M–100M KRW / Mid-market: 200M–300M + 100M–200M / Enterprise & public sector: 300M–500M + 200M–300M. Far more cost-effective than Palantir Foundry.