TECHNOLOGY

One technology base,
a growing set
of industries

Every Syndasis product is built on the same technology base.
Today that is four engines and the agent execution layer,
and we keep adding engines as new judgment is needed.
Only the domain rules are swapped in, so a new industry
comes online quickly on a structure already proven.

  THREE PRINCIPLES

Three design principles

Rules built into the engine

IFRS 17, ICS, RBC and IFRS 9; KDS and MIL-STD; the working rules of contracts, purchasing and trade — all of it lives inside the engine. Citing a KDS clause exactly, validating an actuarial assumption against the regulation, checking trade documents against contract terms: only an engine that has internalized the rules can do that. The rules of a new domain sit on the same framework. Today that is insurance, credit, defense and supply chain; the next industry arrives the same way.

AI judges, people confirm

Every output arrives with the clause it rests on, the calculation behind it and a confidence score. Legally or financially binding acts — signing a contract, placing an order, releasing a payment, filing a customs declaration — run only after a person approves, and the whole path is kept as an audit trail. We don't use black-box AI.

Deploy anywhere

One architecture covers SaaS, on-premises and hybrid. A defense security network, a segregated financial network, or a manufacturing or retail cloud where speed of rollout matters — the environment changes, the product does not.

  AI STACK

Syndasis Core Four engines today + the agent layer

① Document Intelligence
  • Native parsing of HWP, PDF, DOCX and scanned OCR
  • bge-m3 embeddings · dense + BM25 hybrid retrieval
  • Cross-encoder reranking
  • sLLM fine-tuning (EXAONE and Llama families, LoRA) · local serving on air-gapped networks
  • Used in · AI-Milspec · Supply Chain AI Agents · Insurance Risk Platform
② Rule & Knowledge
  • Compliance decisions from a rule engine combined with RAG
  • IFRS 17 · ICS · RBC · IFRS 9 · KDS · MIL-STD · working rules of contracts, purchasing and trade
  • Regulatory data quality rulesets (consistency checks and cleansing)
  • Enterprise rule and knowledge store · a framework where the domain is swappable
  • Used in · Insurance Risk Platform · AI-Milspec · Supply Chain AI Agents
③ Quant & Predictive
  • AI proxy models for fast simulation · Monte Carlo
  • LSTM and Transformer + XGBoost stacking ensemble
  • Time-series forecasting of demand, quantity and deadlines · risk scoring
  • Anomaly detection
  • Term-structured risk curves from discrete-time survival analysis · calibration (binomial test by grade, Hosmer–Lemeshow)
  • Used in · Credit Risk Platform Horizon · Insurance Risk Platform · Supply Chain AI Agents
④ Explainable Decision
  • Natural-language rationale generated from SHAP combined with rules
  • Mandatory citation of the governing clause · confidence score
  • Human approval gate · 100% audit trail · learning from approval and rejection history
  • Fairness (DP, EO) and drift (PSI, CSI) shown on screen
  • Used in · Credit Risk Platform Horizon · Insurance Risk Platform
Agent layer · execution
  • Work broken into three levels · mapping what belongs to an agent and what stays with existing systems
  • Multi-agent orchestration · exception handling · replanning
  • Hybrid AI routing · kill switch · stage-by-stage halt · blocking actions beyond granted authority
  • Used in · Supply Chain AI Agents in full · approval gates and audit trail in the other products
Integration · deployment · validation
  • REST and MCP connectors (ERP, SCM, CRM, WMS, customs, CF engines)
  • React + FastAPI · microservices · the same architecture in every deployment
  • Quantitative acceptance testing on F1, MAPE and MRR@5 · reproducibility against a frozen test set · MCP · A2A · aligned with ISO/IEC 42001
  • Reproducibility · out-of-time validation · frozen-lineage audit
  • Measured on · Credit Risk Platform Horizon, reproduction deviation of 1.8e-7 on the frozen build
+ Next Engine

New engines are added as new kinds of judgment are needed. Each one runs on the same embedded rules, the same approval gate and the same audit trail.

  DEPLOYMENT

Deployment architecture

SaaS · Cloud Native
Multi-tenant cloud on Azure. AKS with VMSS autoscaling. A Southeast Asia region keeps APAC latency low.
Azure AKSMSARBACMulti-tenant
On-Premises · Air Gap
One-click deployment with Docker Compose. Runs fully standalone on a single GPU server (RTX 4090 or better) with no internet connection. Extensible to domestic NPUs and private cloud.
Docker ComposeLDAP/TLSAir-gapped
Hybrid
Source data stays inside the company while the judgment and simulation layers run in the cloud — flexible enough for data sovereignty requirements.
Data sovereigntyFor segregated networksFlexible setup
  SECURITY

Security & compliance

Encryption in transit and at rest

TLS 1.3 in transit, AES-256 at rest. Data is protected on every leg.

Role-based access control (RBAC)

LDAP integration and per-role permissions. Tenants, permissions and credentials are kept separate, and integration never moves source data out.

Automated audit trail

Input → judgment → rationale → approval → execution, all recorded automatically. Evidence for an audit is ready on request.

Full air-gap support

Runs fully standalone with no internet connection, meeting defense security network and financial network segregation requirements.