AEGIS-DocIntel / AMDI-OS

Adaptive Mathematical Document Intelligence Operating System

---

Executive Overview

AEGIS-DocIntel / AMDI-OS is a production-grade Pre-LLM Document Intelligence Operating System that converts unstructured documents (PDF, DOCX, XLSX, PPTX, Images, Speech/Audio) into multi-dimensional, synchronized mathematical representations before exporting token-optimized context to downstream AI agents (ChatGPT, Gemini, Claude, DeepSeek, Qwen, local models).

Built on the 10-tuple Master State Space $D = (P, S, G, R, F, M, T, X, H, E)$, AEGIS-DocIntel enforces formal mathematical guarantees (Theorem 6.1 Spatial DAG Acyclicity, Theorem 6.2 Kahn Topological Determinism, Theorem 9.1 $\frac{1}{2}$-Knapsack Bound, and Monotone Submodular Knapsack $(1 - 1/e)$ Approximation Bound).

"A compiler that converts human documents, tables, images, and audio into mathematical objects before any AI model sees them."

---

Architectural Highlights

• 16 MIOS Mathematical Domains: Topology, Spectral, Physics, Information Theory, Graph Theory, Optimization, Tensor, Probability, Statistics, Harmonic Analysis, Computational Geometry, Control Theory, Decision Theory, Dynamical Systems, Linear Algebra, Numerical Analysis.
• Multimodal Ingestion Engine:
  • Documents: PDF (native & scanned OCR), DOCX, XLSX, PPTX, HTML, Markdown, Plain Text.
  • Speech & Audio: WAV, MP3, FLAC, OGG, M4A with Speech-to-Text (STT) transcription, timestamped segments, and speaker diarization.
  • Visual Layout Decomposition: Bounding box region detection (Heading, Paragraph, Table, Figure, Caption, Header, Footer), image sharpness variance scoring $\sigma^2(\Delta I)$, and 128-D visual feature embedding.
• Pre-LLM Security & Compliance Suite:
  • PII Redaction: Regex & NER detection with policy-driven masking (<US_SSN_REDACTED>).
  • Entity Resolution: Fellegi-Sunter probabilistic matching & NetworkX equivalence clustering.
  • Structural Version Diff Engine: APTED tree-edit distance versioning and diff querying.
  • Ingestion Anomaly Gate: IsolationForest outlier detection & prompt-injection adversarial filter.
  • Unit Normalizer: Locale-aware quantity parsing & point-in-time currency conversion.
  • Query Decomposition Pre-Processor: Sub-query dependency DAG parser and executor.
• LLM Token-Optimized Exporter (LLMTokenOptimizedExporter):
  • Compact Markdown (.md): Minimal padding, dense pipe tables, inline citations [Doc:pX], and contextual prefixing.
  • Ultra-Dense Minified JSON (.json): Abbreviated keys (sys, ctx, sum, cits, meta), zero whitespace, and token budget capping.
• 13 UI Software Dashboard Pages: Full backend API contracts & page data models.

---

Master 16 Mathematical Intelligence Domains

Domain Index	Mathematical Domain	Core Formulations & Algorithms	Engine Implementation
1	Topology	Simplicial complexes, Vietoris-Rips filtration, Betti numbers $H_0, H_1, H_2$	src/math_concepts/topology.py
2	Spectral	Graph Laplacian spectrum $L = D - A$, Cheeger inequality expansion	src/math_concepts/spectral.py
3	Physics	Ising model spin Hamiltonian $H(s) = -\frac{1}{2} s^T J s - h^T s$, simulated annealing	src/math_concepts/physics.py
4	Information Theory	Shannon entropy $H(X) = -\sum p(x) \log p(x)$, mutual information, IB value function	src/math_concepts/information_theory.py
5	Graph Theory	Spatial Reading Order DAG (Thm 6.1/6.2), PageRank power iteration, Hypergraph spectral clustering	src/math_concepts/graph_theory.py
6	Optimization	Monotone submodular knapsack coverage ($(1 - 1/e)$ bound), Modified density greedy knapsack	src/math_concepts/optimization.py
7	Tensor	Multimodal CP / Tucker tensor decomposition, higher-order SVD	src/math_concepts/tensor.py
8	Probability	Bayesian posterior updating $P(\theta \mid D) \propto P(D \mid \theta) P(\theta)$	src/math_concepts/probability.py
9	Statistics	Covariance matrix $\Sigma$, Pearson correlation $R$, statistical moments	src/math_concepts/statistics.py
10	Harmonic Analysis	Fast Fourier Transform (FFT), spectral density decomposition	src/math_concepts/harmonic_analysis.py
11	Computational Geometry	Bounding box Graham scan convex hull, Voronoi proximity diagram	src/math_concepts/computational_geometry.py
12	Control Theory	Proportional-Integral-Derivative (PID) error stability feedback loop	src/math_concepts/control_theory.py
13	Decision Theory	Expected utility hypothesis, Minimax regret decision matrix	src/math_concepts/decision_theory.py
14	Dynamical Systems	Phase space trajectory, largest Lyapunov exponent estimation	src/math_concepts/dynamical_systems.py
15	Linear Algebra	Singular Value Decomposition ($A = U \Sigma V^T$), low-rank approximation	src/math_concepts/linear_algebra.py
16	Numerical Analysis	Matrix condition number $\kappa(A) = |A| |A^{-1}|$, floating-point error bounds	src/math_concepts/numerical_analysis.py

---

13 UI Software Dashboard Pages Suite

Page Index	UI Dashboard Module	Primary System Functionality	File Path
1	Upload Dashboard	File ingestion progress tracking, file type validation & error reporting	ui/src/pages/upload_dashboard.py
2	Document Explorer	Interactive document browsing, layout tree navigation & metadata filtering	ui/src/pages/document_explorer.py
3	Geometry Dashboard	Spatial bounding box coordinates $[x, y, w, h]$ & spatial reading order DAG	ui/src/pages/geometry_dashboard.py
4	Matrix Dashboard	Multi-table structure extraction, financial statistical metrics & unit normalization	ui/src/pages/matrix_dashboard.py
5	Graph Dashboard	Node degree, closeness, betweenness centrality, PageRank & hypergraph clustering	ui/src/pages/graph_dashboard.py
6	Memory Dashboard	L0–L5 multi-tier hierarchical memory cache monitoring	ui/src/pages/memory_dashboard.py
7	Retrieval Dashboard	Hybrid 7-method vector + BM25 + visual ColPali search interface	ui/src/pages/retrieval_dashboard.py
8	Analytics Dashboard	Cross-document entity resolution & analytical insights	ui/src/pages/analytics_dashboard.py
9	Performance Dashboard	Sub-second engine latency, memory allocation & throughput metrics	ui/src/pages/performance_dashboard.py
10	Agent Dashboard	AI connector management (ChatGPT, Gemini, Claude, DeepSeek, Qwen)	ui/src/pages/agent_dashboard.py
11	Settings Dashboard	System-wide configuration, storage backends & environment flags	ui/src/pages/settings_dashboard.py
12	Math & Advanced Dashboard	PII redaction, entity canonicalization, version diff, anomaly gate & 16 MIOS domains	ui/src/pages/math_advanced_dashboard.py
13	Speech & Image Dashboard	Audio STT transcription, speaker diarization, SNR & visual image layout parsing	ui/src/pages/speech_image_dashboard.py

---

Installation & Verification

Prerequisites

• Python 3.12+
• Dependencies listed in requirements.txt / requirements.lock.txt

1. Clone & Environment Setup

git clone https://github.com/SahilKhutey/AEGIS-DocIntel.git
cd AEGIS-DocIntel
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate
pip install -r requirements.txt

2. Run Test Suite

To run all 860+ passing unit test items:

python -m pytest tests/

3. Start REST API Server

python -m uvicorn src.main:app --host 0.0.0.0 --port 8000 --reload

Access interactive OpenAPI documentation at http://localhost:8000/docs.

---

REST API Endpoints Overview

HTTP Method	Route	Description
POST	/v1/documents/upload	Upload & ingest document (PDF, DOCX, XLSX, PPTX, Image, Audio)
POST	/v1/query/	Execute hybrid RAG query over ingested context
POST	/v1/advanced/compliance/pii-scan	Scan & redact PII with policy rules
POST	/v1/advanced/entity/resolve	Resolve cross-document entity mentions into canonical clusters
POST	/v1/advanced/versioning/diff	Compute structural tree-edit distance between document versions
POST	/v1/advanced/ingestion/anomaly-check	Scan document for IsolationForest outliers & prompt injection
POST	/v1/advanced/matrix/normalize-quantity	Parse locale quantities & normalize currencies
POST	/v1/advanced/query/decompose	Decompose complex query into sub-query dependency DAG
POST	/v1/advanced/math/unified-evaluation	Evaluate document state $D$ across all 16 mathematical domains
POST	/v1/advanced/ingestion/parse-speech	Transcribe speech audio with speaker diarization & SNR scoring
POST	/v1/advanced/ingestion/parse-image-layout	Decompose document image layout & compute sharpness score
POST	/v1/advanced/export/llm-optimized	Export token-optimized Markdown or minified JSON context

---

Repository Structure

AEGIS-DocIntel/
├── src/
│   ├── ael/                   # Adaptive Export Layer & token budget allocator
│   ├── api/                   # FastAPI routes, auth, schemas, and routers
│   ├── compliance/            # PII detection & policy redaction engine
│   ├── connectors/            # AI agent connectors (LangChain, LlamaIndex, Claude, etc.)
│   ├── core/                  # DocumentObject, Master State D, AMDIOrchestrator
│   ├── engines/               # 12 core mathematical & spatial reading order engines
│   ├── entity/                # Cross-document entity resolution (Fellegi-Sunter)
│   ├── export/                # Exporters (Markdown, JSON, YAML, LLMTokenOptimizedExporter)
│   ├── ingestion/             # PDF, DOCX, XLSX, PPTX, Image, SpeechLoader, OCREngine
│   ├── math_concepts/         # MasterUnifiedMathEngine & 16 mathematical domains
│   ├── query/                 # Query decomposition DAG pre-processor
│   ├── services/              # ServiceContainer dependency injection
│   └── versioning/            # Structural diff engine (APTED tree-edit distance)
├── ui/
│   └── src/pages/             # 13 UI Dashboard software pages
├── tests/                     # 860+ passing pytest test items
├── Aegis Doc/                 # 11 Foundational Publications & Monographs
└── requirements.txt           # Dependency requirements

---

License & Authorship

Sahil Khutey (with AI Research Collaborator, Gensouls Lab)
July 2026 Monograph Series & System Specifications.
Proprietary — All rights reserved.