Development of a Hybrid Intelligence Decision Support System for Strategic Planning Support

Hybrid Intelligence Decision Support System for evidence-based strategic planning under information overload and uncertain signals.
It evaluates strategic plans against a transparent, traceable evidence base, integrates human strategic judgment throughout the workflow, and prepares results for strategic assessment.

Table of Contents

1. Overview
2. Technical innovation
3. Built With
   • Deployment
4. Injecting the hybrid intelligence into the GE-McKinsey matrix
5. Workflow screenshots
6. Frontend
7. Backend
8. Additional docs

---

Overview

The system pairs a multi-step frontend workflow with a FastAPI backend that manages strategy stages, evidence scoring, and NLI outputs.

Technical innovation

Solution Objectives (Hevner et al., 2004)

Built With

Frontend

Backend

Deployment

Injecting Hybrid Intelligence into the GE-McKinsey Matrix: A Data-Driven Decision Support Extension

Scoring Logic (Implementation of the Solution Objectives)

The following part is intentionally reduced to the conceptual structure and core stages. Detailed technical implementation and parameterization are documented in the project folder.

1. Preprocessing
   The evidence base (for example forecast and risk reports in PDF format) is extracted, cleaned, split into chunks, and indexed as embeddings in a vector database together with source-related metadata. If needed, a language model converts content from charts or tables into NLI-compatible premises.

2. Hypothesis generation
   Strategic plans are transformed into declarative, NLI-compatible hypotheses. This enables consistent semantic matching with the vector database and NLI models.

3. Information retrieval for candidate generation
   The evidence base is reduced to relevant candidates. Relevance is determined with vector-based cosine similarity on L2-normalized representations.

4. NLI-based pair scoring
   Relevant evidence candidates are evaluated pairwise against the strategic hypothesis (support, contradiction, neutral). The resulting softmax scores quantify the strength of the semantic relation.

5. Model-based evaluation (AI baseline)
   For each dimension (forecast and risk, analogously), the mean and population variance of included evidence are aggregated.

   Let S_F = {s_1, ..., s_n} be the set of included forecast evidence scores with s_i in [0,1].

   

   Based on mu_F, sigma_F^2, and n, a normal-approximation uncertainty interval (z = 1.96) is computed and bounded to [0,1]:

   

6. Hybrid score fusion
   The alignment slider a_F in [0,1] shifts the AI baseline score:

   

   The stability slider c_F in [0,1] scales the interval width:

   

   The output of this stage is the hybrid mean score mu_F* and the hybrid interval [L_F*, U_F*], both constrained to [0,1].

7. Strategy derivation via area shares
   The interval zones are placed as a rectangle in the GE-McKinsey matrix. For each cell (r,c), overlap area is calculated relative to the total overlap:

   

   The final output is a percentage-based Strategy Distribution across all matrix cells.

Workflow screenshots

Step 1 - Strategic Plan

Define the strategic intent, segment, and region, or select a predefined plan.

Step 2 - Evidence Selection

Curate the knowledgebase by including or excluding evidence, with score guidance.

Step 3 - Evidence Reasoning

Adjust alignment and stability sliders to reflect the evidence signal.

Step 4 - Evidence Positioning

Place evidence in the GE-McKinsey matrix and refine the final position.

Step 5 - Recommendation Dashboard

Review strategic insights and the evidence foundation behind the recommendation.

Feature detail

Source document viewer for evidence traceability and report inspection.

Frontend

React + Vite frontend for the Hybrid Intelligence Decision Support system. Focus: multi-step evaluation workflow, evidence curation, and visualization.

Local run

Requirement: Node.js >= 18

npm install
npm run dev

Project structure

.
|-- public/              # static assets
|-- src/
|   |-- main.tsx         # React root, router, providers
|   |-- router/          # route definitions
|   |-- pages/           # screen-level routes
|   |-- layouts/         # app layout shells
|   |-- components/      # shared UI components
|   |-- features/        # workflow features (charts, selection, pdf)
|   |-- data/            # datasets and matrix presets
|   |-- lib/             # API client, helpers
|   |-- hooks/           # UI hooks
|   |-- styles/          # Tailwind entrypoint
|   `-- types/           # type declarations
|-- index.html
|-- package.json
|-- vite.config.ts
|-- tailwind.config.js
|-- tsconfig*.json
`-- vercel.json

Each top-level folder represents a system responsibility.

Architecture

Workflow view

+====================================================================+
||......................:: 1-Strategic Plan ::......................||
||                                                                  ||
||                        strategic-plan.tsx                        ||
||                                                                  ||
+====================================================================+
                                   v
+====================================================================+
||....................:: 2-Evidence Selection ::....................||
||                                                                  ||
||     evidence-selection-loading.tsx -> evidence-selection.tsx      ||
||                                                                  ||
+====================================================================+
                                   v
+====================================================================+
||....................:: 3-Evidence Reasoning ::....................||
||                                                                  ||
||                      evidence-reasoning.tsx                      ||
||                                                                  ||
+====================================================================+
                                   v
+====================================================================+
||...................:: 4-Evidence Positioning ::...................||
||                                                                  ||
||   evidence-positioning-loading.tsx -> evidence-positioning.tsx    ||
||                                                                  ||
+====================================================================+
                                   v
+====================================================================+
||.................:: 5-Recommendation Dashboard ::.................||
||                                                                  ||
||                   recommendation-dashboard.tsx                   ||
||                                                                  ||
+====================================================================+

API

App entry point:

http://127.0.0.1:5173

Examples:

http://127.0.0.1:5173/login
http://127.0.0.1:5173/strategic-plan
http://127.0.0.1:5173/dashboard

Backend

FastAPI backend for the Hybrid Intelligence Decision Support system. Focus: orchestrating strategy workflows, NLI stages, and scoring via HTTP.

Local run

Requirement: Python >= 3.12

python -m venv .venv
source .venv/bin/activate

python -m pip install --upgrade pip poetry
poetry install

poetry run serve

Project structure

.
|-- app/
|   |-- __main__.py        # uvicorn entry point
|   |-- main.py            # app factory, CORS, health
|   |-- api/               # HTTP routes (v1)
|   |-- modules/           # hybrid services and schemas
|   |-- pipelines/         # NLI stage scripts
|   |-- infrastructure/    # paths, persistence, presets
|   |-- config/            # settings
|   |-- scripts/           # CLI helpers
|   `-- data/              # default data/workdir root
|-- pyproject.toml
|-- poetry.lock
|-- pytest.ini
`-- README.md

Each top-level folder represents a system responsibility.

Architecture

Stages view

+====================================================================+
||..................:: Preprocessing artifacts ::...................||
+====================================================================+
                                   v
+====================================================================+
||........................:: 2-Hypothesen ::........................||
||                                                                  ||
||                      strategy_hypotheses.py                      ||
||                                                                  ||
+====================================================================+
                                   v
+====================================================================+
||........................:: 3-Embeddings ::........................||
||                                                                  ||
||                 forecast-reports | risk-reports                  ||
||                                                                  ||
+====================================================================+
                                   v
+====================================================================+
||.......................:: 4-PremisePairs ::.......................||
||                                                                  ||
||                 forecast-reports | risk-reports                  ||
||                                                                  ||
+====================================================================+
                                   v
+====================================================================+
||.........................:: 5-Reports ::..........................||
||                                                                  ||
||                          merge_pairs.py                          ||
||                                                                  ||
+====================================================================+
                                   v
+====================================================================+
||........................:: 6-UserReview ::........................||
||                                                                  ||
||                        add_user_status.py                        ||
||                                                                  ||
+====================================================================+
                                   v
+====================================================================+
||.........................:: 7-Scoring ::..........................||
||                                                                  ||
||                 score_summary.py | intervall.py                  ||
||                                                                  ||
+====================================================================+
                                   v
+====================================================================+
||.....................:: 8-HumanKalibration ::.....................||
||                                                                  ||
||                       human_calibration.py                       ||
||                                                                  ||
+====================================================================+
                                   v
+====================================================================+
||.........................:: 9-Strategy ::.........................||
||                                                                  ||
||                     compute_distribution.py                      ||
||                                                                  ||
+====================================================================+

Responsibilities

• API layer: FastAPI app, v1 router, health endpoints.
• Service layer: modules/hybrid/services orchestrate workflow state and stage execution.
• Stages: app/pipelines/nli scripts for retrieval, NLI pairing, reports, scoring.
• State handling: workdir files under NLI_DATA_ROOT/NLI_WORKDIR and presets in PRESETS_DIR.
• Configuration: pydantic-settings loads .env and environment variables.

Configuration

Environment variables are loaded from .env in the repository root (if present).

Name	Responsibility	Default
APP_HOST	Bind address	127.0.0.1
APP_PORT	HTTP port	8000
APP_RELOAD	Auto reload	false
FRONTEND_ORIGINS	CORS allowlist	http://localhost:5173, https://hybridintelligence.dev
OPENAI_API_KEY	OpenAI API key for stage scripts	unset
NLI_DATA_ROOT	Base data directory	app/data/nli
NLI_WORKDIR	Stages workdir	NLI_DATA_ROOT/workdir
PRESETS_DIR	Preset files	presets
NLI_MODEL_NAME	Local NLI model	microsoft/deberta-large-mnli

API

OpenAPI schema:

http://127.0.0.1:8000/openapi.json

Examples:

curl http://127.0.0.1:8000/health
curl http://127.0.0.1:8000/hybrid/pipeline/status

Additional docs

• frontend/docs_frontend.md
• backend/docs_backend.md