AI Orchestrator

A Specialized and Secure AI Orchestrator for Swiss Financial Compliance

View the Project on GitHub Digital-AI-Finance/wecan-innosuisse-ai-draft

Project Goals & Success Metrics


Overview

This page defines the quantifiable objectives for Innosuisse Innovation Project 133.672 IP-SBM. Each objective has a measurable target, a defined baseline, and a specific measurement method tied to a milestone.


Quantifiable Objectives

OBJ1: Document Extraction Accuracy

Attribute Value
Target 90% accuracy on 50-100 page compliance documents
Baseline Current AI: 95%+ on single pages, 70-80% on extended documents
Measurement Blind assessment on held-out test set
Milestone MS5 (M20 - August 2027)
Work Packages WP2, WP3

OBJ2: Zero-Shot Schema Mapping

Attribute Value
Target F1 > 85% across 50 enterprise CRM schemas
Baseline No prior zero-shot mapping capability exists
Measurement Evaluation against Salesforce, SAP, MS Dynamics, HubSpot, custom platforms
Milestone MS4 (M16 - April 2027)
Work Packages WP4

OBJ3: Hallucination Reduction

Attribute Value
Target 40% reduction from baseline
Baseline 23.5% total hallucination rate
Target Rate 14% or lower
Measurement Controlled comparison against baseline model
Milestone MS3 (M12 - December 2026)
Work Packages WP2

OBJ4: Processing Time

Attribute Value
Target < 2 hours per 100-page document
Baseline 2-3 weeks manual processing
Measurement End-to-end timing benchmark
Milestone MS5 (M20 - August 2027)
Work Packages WP3, WP5

OBJ5: Institutional Deployments

Attribute Value
Target 3-5 Swiss financial institutions
Baseline 0 deployments
Measurement Production use for minimum 1 month, 50+ documents per institution
Milestone MS5 (M20 - August 2027)
Work Packages WP4, WP5

OBJ6: Technology Readiness Level

Attribute Value
Target Advance from TRL 3 toward TRL 5-6
Baseline TRL 3 (experimental proof of concept)
Measurement Evidence documentation portfolio
Milestone MS5 (M20 - August 2027)
Work Packages All

OBJ7: On-Premise Deployment

Attribute Value
Target 7-13B parameter models deployable on-premise
Baseline Cloud-only LLM deployments
Measurement Cost comparison vs cloud solutions
Milestone MS4 (M16 - April 2027)
Work Packages WP2, WP4

OBJ8: Multilingual Validation

Attribute Value
Target 500 documents across 4 languages
Baseline 0 validated multilingual documents
Measurement Corpus: 40% DE, 30% FR, 15% IT, 15% EN
Milestone MS3 (M12 - December 2026)
Work Packages WP2, WP3

Success Metrics Summary

Dimension Metric Target Validated At
Accuracy Document extraction 90% MS5
Speed Processing time < 2 hours MS5
Scale Deployments 3-5 institutions MS5
Model Parameters 7-13B on-premise MS4
TRL Advancement Toward TRL 5-6 MS5
Data Annotated documents 500+ MS3
Hallucination Reduction 40% MS3
Languages Coverage DE/FR/IT/EN MS3
Schema Zero-shot mapping F1 > 85% MS4

Source: Project Inventory Section 3, Innosuisse Application 133.672 IP-SBM