DOMAIN-NATIVE AI AGENTS · REGULATED & COMPLEX ENTERPRISE WORK

Complex rules,
solved by agents.

Insurance closing, defense specification review, contracts, purchasing and trade —
Syndasis implements expert judgment through AI agents that carry each industry's rules inside them. They show the reasoning, and act only under your team's approval.

Request a demo Explore the technology
4 Domains Industries with the rules built in
1 Core One technology base across every product
100% Rationale · approval · audit trail
Insurance Risk Platform
Automated IFRS 17, ICS and RBC assumption setting, validation and P&L analysis. Connects to your CF engine without replacing it.
Credit Risk Platform Horizon
Term-structured PD curves with written rationale. Three-band routing, never an automatic decline.
AI-Milspec
AI analysis of defense and public specification documents. Full KDS, MIL-STD and HWP support. Runs on-premises, air-gapped.
Supply Chain AI Agents
Five agents across contracts, purchasing, sourcing, trade and customer support. Approval-based execution, with no ERP or SCM replacement.
Venture company
Korea Fintech Center · digital finance technology program
Kumoh National Institute of Technology · joint R&D
AI+ OpenData Challenge
2 patents filed
Partners NTT Data Thailand Kumoh National Institute of Technology
  Our Focus

Four
platform products

Work where rules govern the decision, and work where exceptions do. Two kinds of complexity a general-purpose AI cannot reach, solved by four products on one technology base.

Insurance Risk
Insurance Risk Platform
Eight modules automate IFRS 17, ICS and RBC work end to end: assumption mapping, calculation, validation and P&L analysis. Your existing CF engine is connected by API rather than replaced, and it doesn't stop at the number — it delivers the reasoning behind it and product-level P&L analysis.
Assumption AI Hub— an automated assumption setting and validation workflow
AI Proxy Simulator— 12 hours → 1 minute, 1,000× faster simulation
AI Attribution Analyzer— automatic decomposition and explanation of P&L and capital movements
Credit Risk
Credit Risk Platform Horizon
New underwriting, ongoing monitoring, additional-loan assessment and model validation all run on one platform. Instead of a single PD at drawdown, it produces a risk curve over the life of the loan, and every decision leaves a written rationale and a record of who confirmed it. Your existing credit systems stay where they are.
Term-structured PD curve— it answers when, not just how much
A written rationale per decision— the number and the explanation share one axis
Three-band routing— a lighter review load, never an automatic decline
Defense & Public
AI-Milspec
Military and public specification documents that cannot leave the network are analyzed by an sLLM-RAG running air-gapped. Native HWP parsing with hybrid retrieval and reranking extracts clauses precisely, automating gap analysis, RTM and compliance audit.
2–4 expert-weeks → tens of minutes— gap analysis and specification matching
Full HWP support— the format of 90% of Korean military documents
No internet connection required— on-premises, air-gapped
Supply Chain
Supply Chain AI Agents
We broke contracts, purchasing, sourcing, trade and customer support into unit tasks and mapped which belong to an agent and which stay with your ERP or SCM. Agents handle the repetitive work that needs judgment and show their reasoning; once a person approves, the result is written back to your existing systems.
Five agents working together— contracts, purchasing, sourcing, trade, customer support
Document checks · risk alerts · rule checks— the shared intelligence underneath
ERP · SCM · CRM · WMS · customs— integrated, not replaced; execution after approval
  How It Works

More products,
a deeper core

Different domains, same mechanics. On top of engines for document intelligence, embedded rules, quantitative forecasting and rationale, AI judges, a person confirms, and your existing systems execute.

01
Reading documents and data
Contracts, specifications, customs paperwork and closing data are read as they are — HWP, PDF, DOCX, Excel — alongside connected ERP, SCM and CRM data. No preprocessing, no format conversion.
02
Deciding from rules and forecasts, with the reasoning attached
Depending on the task, the work is routed to a rule engine, RAG, ML or an LLM. Nothing is left to a single LLM, and every decision arrives with its governing clause, its calculation and a confidence score.
03
Approval → execution in your systems → recorded
Binding acts — signing a contract, placing an order, releasing a payment, filing a customs declaration — are approved by a person. Approved results flow into your ERP, WMS and CF engine, and the whole path is kept as an audit trail.
document-ingestion · processing
PDFCommercial_Invoice_IMP-2026-0912.pdf1.2 MB
HWPpurchase_agreement_BrandA_v3.hwp2.4 MB
HWPKDS-0050-0027_specification_v3.hwp2.1 MB
DOCXIFRS17_assumptions_2026Q2.docx892 KB
hybrid-routing · rule + rag + ml
Trade agent · Document AI → Rule + RAG
Document cross-check — invoice vs PO
Qty 1,000 vs ordered 980 → mismatch
Basis: purchase agreement §3.2, ±1% tolerance exceeded → partial receipt recommended
Customs requirement check
C/O origin marking valid · 2 HS code candidates · no missing documents
confidence
93%
human-approval · execution · audit-trail
Proposed decision — partial receipt, 980 EAawaiting approval
Partial receipt confirmed · excess returnedapproved
ERP purchase voucher · WMS receipt postedexecuted
Draft notice to the supplier (customer support agent)auto-drafted
Audit trail100% recorded
Approval and edit historyfed back into the rules
  Why Syndasis

Why Syndasis

In the places a general-purpose AI service structurally cannot reach,

these are the four principles we hold to.

01
We build the domain's rules into the engine
Paste a contract into ChatGPT and you get a plausible summary. But validating actuarial assumptions against IFRS 17, ICS and RBC, citing a KDS clause exactly, or checking trade documents against contract terms takes an engine with the rules inside it. Because only the domain changes within the same framework, a new industry comes online fast.
02
AI judges. People decide.
Regulators, auditors and executives do not accept black boxes. Every output carries its governing clause, its calculation and a confidence score, and legally or financially binding acts — signing a contract, confirming an order, releasing a payment, filing a customs declaration — run only after a person approves.
03
We don't replace your existing systems
An insurer's CF engine, a defense contractor's document system, a company's ERP, SCM, CRM, WMS and customs systems — all of it stays. Syndasis adds only the judgment, drafting and exception-handling layer on top. Switching cost and internal resistance stay low.
04
Air-gapped or in the cloud
One architecture covers SaaS, on-premises and hybrid. From defense security networks and segregated financial networks to manufacturing and retail clouds where rollout speed matters, the same product line handles it.
  Use Cases

When you would use it

Four situations where Syndasis products actually run.

Supply Chain AI Agents · multi-agent replanning
A stockout is coming.
Purchasing, contracts, sourcing and customs have to be decided together.
One person moves between ERP, WMS, contracts, email and customs systems, checking order quantity, contract terms, delivery dates and document requirements one at a time. If customs holds the shipment, the review starts over.
→The purchasing agent reads sales, inventory and inbound data to decide whether, how much and when to order; the contract agent checks it against the timing, quantity and unit-price terms agreed with the brand's head office
→The sourcing agent analyzes supplier lead times and supply risk; the trade agent pre-verifies customs requirements, required documents and HS code candidates. Every decision comes with its reasoning
→A person approves , then the order is posted to ERP and WMS. If customs puts a hold on it, the sourcing agent proposes alternate sourcing or a split shipment and the customer support agent identifies the affected orders and customers and drafts the notice
Pilot targets: 50% less time verifying trade documents · 70% task automation · 90% successful replanning on exceptions. Rationale, approval and history recorded in full.
AI-Milspec · Defense Specification Analysis Platform
Does the delivery document meet
the governing specification?
One expert manually matches hundreds of KDS clauses against the delivery document. Two weeks, with a standing risk that something slips through. And the documents cannot leave the network.
→The delivery document and the governing specification are parsed automatically by AI inside the closed network (HWP, PDF, DOCX and more)
→For each requirement, an automatic four-way verdict — met, not met, partially met, not applicable plus the governing clause and a confidence score
→An audit report with the overall compliance rate, prioritized gaps and recommended actions is generated automatically. The final verdict is confirmed by the responsible engineer
Result: Work that took an expert 2–4 weeks now takes tens of minutes. Citation accuracy above 90%, running fully standalone on a closed network.
Insurance Risk Platform · Assumption AI Hub + AI Proxy Simulator
The assumptions are set,
but validation and P&L analysis still live in Excel.
The CF engine finishes the calculation, but mapping, validating and approving assumptions — and explaining what moved P&L — is still manual. The audit trail breaks somewhere between spreadsheets.
→EngineConnect connects to CF engine output from Moses, Prophet and others by API, non-invasively (no engine replacement); FDP validates and corrects the source data
→Assumption AI Hub automates the assumption mapping → calculation → validation → approval workflow; AI Proxy Simulator recalculates a 12-hour scenario in under a minute
→AI Attribution Analyzer decomposes what moved P&L and capital using XAI; Auto-Report generates regulatory-format reports and the audit trail automatically
Result: Closing lead time from 3–4 weeks to T+1. 90% less time writing reports. Audit evidence prepared automatically.
Credit Risk Platform Horizon · lifetime PD + written rationale
A first-time small-business applicant.
The decision and the reason behind it both have to be recorded.
A credit score and a single point-in-time PD do not reveal that the risk piles up right when the grace period ends. For an applicant with a thin file, a missing value becomes a deduction, and the reason for the decision has to be written up separately.
→A discrete-time survival model with an XGBoost ensemble produces marginal and cumulative PD curves over the life of the loan, plus a fine-grained grade. A thin file is treated not as a deduction but as a confidence adjustment.
→A cost-weighted threshold routes each case into three bands — fast-track, review, deep review —and on the same SHAP axis that produced the score, three to five sentences of rationale are generated
→A person confirms the outcome. There is no automatic decline. When a decision differs from the system's, the reason is recorded, and the applicant receives a notice that includes the rationale
Validation discipline: Out-of-time splits only, every random seed fixed, frozen-build reproduction deviation of 1.8e-7. Fairness (DP, EO) and drift (PSI, CSI) are shown on screen, not buried in a document.
Ready to Transform?

Where rules and exceptions are complex,
AI agents make the difference

See which judgments in your operation an agent can take on —
in a 30-minute demo.

Request demo

Reply within one business day · no sign-up required