RepoScroller - ALCOA+ Document Intelligence & Deduplication

Active Storage Mounts:

Probing storage roots...

🏷️ Unique Documents

Single source of truth

📁 Physical Locations

Distributed files mapped

🕒 Version Links

Evolved & draft chains

📄 Deduplicated Savings

Exact duplicate copies (click to filter)

🧠 Knowledge Base

Active

Vectors & Graph Nodes

🔄

The Process Scroller: 5-Stage Digitalization & Provenance Pipeline

Real-time telemetry and activity tracking across Crawler, Chunker, Cascaded Analyzer, CUDA Embedder, and Lineage Graph Engine

Pipeline Active

📂

Storage Ingest & Crawler

crawler.py

Watching

MultiRootCrawler scanning distributed mounts with antichronological priority and continuous PollingObserver.

Configured Roots: 6 Mounts

Online Roots: 6/6 Accessible

Scan Order: Antichronological

Observer Status: Ready

➔

📄

Context Chunker & PDF I/O

chunker.py + telemetry.py

512 Tokens

DocumentChunker sliding window tokenization with context header injection: [Doc | Cat | Date | Src].

Window / Overlap: 512 / 64 tokens

PDF / Doc I/O Speed: 0.0 MB/s

Files Processed: 11,202 docs

Avg Extraction: -- ms

➔

🤖

Cascaded LLM Analyzer

analyzer.py

Cascaded

DocumentAnalyzer: 3-tier cascade (OpenRouter ➔ Ollama Llama 3.2 ➔ Heuristic Rules) for taxonomy & maturity.

Active LLM Target: PC2 (3B CUDA)

Taxonomy Categories: 18 Slugs

Lifecycle Scored: Final / Draft / Review

Auto-Taxonomy: Active Discovery

➔

⚡

CUDA Embedding Engine

sidecar_worker.py

RTX 3060

KnowledgeBaseSidecarWorker async HTTP pool (6x) to PC2 NVIDIA RTX 3060 CUDA GPU.

Embedding Model: snowflake-arctic-embed2

Vector Dimensions: 1024-dim Floats

GPU VRAM Allocated: 664 MB Resident

Worker Pool: 6x Async HTTP

➔

🕒

Lineage & WAL Graph Store

repository.py + graph_store.py

WAL Mode

ACID document ledger with version lineage relationships, duplicate deduplication, and Property Graph indexing.

Version Links: 3,555,530

Graph Entities: 14,728 Nodes

Graph Relationships: 1,066,018 Edges

Dedup Savings: 8,177 Files

🧠

Dual-Node Realtime Ollama Process Inspector (`/api/ps`)

Direct live telemetry on model residency in memory, VRAM allocation, context windows, quantization, and expiration countdowns

⟳ Synced: --:--:--

💻

PC1 Host Engine (CPU / Host RAM)

http://127.0.0.1:11434/api/ps ↗

Online

Model	Family / Params	Quant	Context	Residency	Expires In
Probing http://127.0.0.1:11434/api/ps...

3 models loaded • ~15.4 GB Host RAM • 100% CPU

⚡

PC2 Remote CUDA Node (NVIDIA RTX 3060)

http://nitro-an51755:11434/api/ps ↗

⚡ CUDA Ready

Model	Family / Params	Quant	Context	GPU VRAM	Expires In
Probing http://nitro-an51755:11434/api/ps...

1 model in VRAM (664 MB) • 100% GPU Accelerated

📡

Pulsed Realtime Telemetry & Ingestion Hub

Continuous live heartbeats from dual Ollama engines, KB Sidecar, and Diagnostic Event stream

PC1 (127.0.0.1:11434) 3 models ↗ PC2 (nitro-an51755:11434) 664MB CUDA ↗ Sidecar /stats 87.7k chunks ↗ Diagnostics /logs Active Stream ↗

⚡ Sidecar Knowledge Base Ingestion Idle (Standby)

Processed: 11,202 / 25,137 docs (44.5%) • Indexed: 87,721 vectors • Graph: 14,728 nodes

⚡ 0.0 files/m • 0.0 chunks/m

⚡ Split Workload Architecture ⚡ CUDA Remote Node (PC2) snowflake-arctic-embed2:latest

api/chat:

⟳ Synced: --:--:--

💻

PC1 Host Engine api/chat

http://127.0.0.1:11434

Probing...

Loaded:

Model (CPU) Loading...

Local Ping --

Chat Calls

0 calls

Processor / RAM

100% CPU

Host Status / Context

idle

⇄

LAN Stream & Throughput PC1 ⇄ PC2

http://localhost:11434 ⇄ http://NITRO-AN51755:11434

Active Stream

Text Chunks ➔ PC2 CUDA POST /api/embed

Dense Vectors ➔ PC1 SQLite 1024-dim Floats (200 OK)

Producer: Idle

➔

Embed Q: 0/100

➔

HTTP Pool: 6x (0 active)

➔

DB Q: 0/100

➔

DB Writer: Idle

➔

KG Validator: Idle

📄 PDF I/O: 0.0 MB/s (0 f/m)

🌐 LAN Ping: -- ms PC1 ⇄ PC2

💾 WAL DB: -- MB Healthy

🤖 Chat LLM: PC1 (3B) 0 calls

🛡️ KG Auditor: Idle Certified

Ingestion Rate -- files/min

Chunk Velocity -- chunks/min

Token Batch

1,206 tokens

Checkpoints

0 (WAL Normal)

Vector Payload

37.702 MiB

⚡

PC2 CUDA Server (RTX 3060) api/embed

http://NITRO-AN51755:11434

Probing...

VRAM Loaded:

Embedding Model snowflake-arctic-embed2

LAN Ping --

Vectors Embedded

0 chunks

VRAM Loaded

VRAM Standby

CUDA & Context

100% GPU

CPU Load -- %

Host RAM -- GB

GPU Compute -- %

VRAM (6GB Max) -- MB (0%)

🕒

Version Lineage & Document Evolution Explorer

Single Source of Truth: Automated version provenance, evolutionary derivation chains, and duplicate savings

🔗 Lineage Links: 3,555,530

📄 Duplicates Saved: 8,177

✅ Final: 1,504

📝 Drafts: 716

Parent Document (Predecessor)	Relationship	Child Document (Successor)	SimHash Score	Maturity Delta	Status Transition	Actions
Loading version lineage relationships...

🧬 Sample Evolutionary Chain (Provenance Graph) SimHash Cosine > 96%

v1 • DRAFT

260909_0813_IngestionReport.md

Maturity: 0.320 • SimHash 96.8%

derived_from ➔

v2 • FINAL

260909_1414_IngestionReport.md

Maturity: 0.850 • SimHash 98.4%

supersedes ➔

v3 • LATEST FINAL

260909_1619_IngestionReport.md

Maturity: 0.980 • Single Source of Truth

🩺

Real-Time Telemetry & Diagnostic Event Stream

Live pulsed activity log from /api/v1/diagnostics/logs across Crawler, Chunker, Analyzer, Sidecar, and CUDA GPU

Auto-scroll

Diagnostic event stream listening on http://127.0.0.1:8090/api/v1/diagnostics/logs...

Document Ledger (WAL)

Document ↕	Category ↕	Maturity ↕	Status ↕	Copies ↕	Doc Date ↕	Reception ↕	Due Date ↕	Actions
Loading document ledger...

🧠

Multi-Signal GraphRAG Sovereign Explorer

Deep hybrid retrieval combining dense RTX 3060 CUDA vector embeddings, BM25 keyword matching, and 1-hop Property Knowledge Graph expansion with Reciprocal Rank Fusion (RRF).

⚡ NVIDIA RTX 3060 CUDA Accelerated

🔍

Quick Prompts:

⚡ Sidecar Indexing Pipeline snowflake-arctic-embed2:latest

Ingestion Progress 0%

Completed: 0 Active: 0 Queue: 0 Failed: 0

0 Chunks Embedded

0 Docs Processed

Pipeline Idle

Session: 0 docs • 0 chunks

🕸️ Property Knowledge Graph Hybrid Neo4j/SQLite

0 Entities (Nodes)

0 Relations (Edges)

0 Document Links

Entity Distribution

Click type to inspect entities

📊 Retrieved Candidates & Signal Breakdown 0 candidates

Limit:

🔬

Ready for Multi-Signal Retrieval

Enter a query above or click one of the quick prompts to execute dense vector search on PC2 CUDA, BM25 scoring, and graph entity neighborhood expansion.

Data Science

3D Knowledge Graph Universe

Topological Clustering & Visual Analytics

Dim: Thematic Gravitational Space Algo: Thematic Solar Centroids (k=7) Algo: Unsupervised K-Means Topology: 1,000 Nodes • 3,000 Links Renderer: Three.js WebGL GLSL

📐 Axes of Importance & Clusters

X-Axis

Domain Specificity & Category Divergence PCA-1 • Latent Semantic Variance

Y-Axis

Temporal Recency & Lifecycle Maturity PCA-2 • Historical Lineage & Validation

Z-Axis

Graph Centrality & Hub Authority Degree PCA-3 • Network Eigenvector Density

Topical Clusters

Quota: 250 max / type

💬

GraphRAG-Powered AI Interrogation Assistant

Verified multi-document reasoning grounded in vector embeddings, knowledge graph entities, and ALCOA+ ledger records via the selected reasoning node.

🧠 GraphRAG Grounded

Hello! I am your RepoScroller Sovereign Interrogation Assistant.

I can analyze version lineages, search for duplicate drafts, and answer complex cross-repository inquiries grounded in your Knowledge Graph and Vector Store.

3D Author-Candidate–Document Graph

This graph includes raw person/document associations for context. Only explicit, high-confidence authorship evidence appears in the human review queue below.

Approved Pending review Declined Document 10+ distinct linked documents flags review only; it is not credibility or authorship proof and never auto-approves.

Loading author graph…

Author candidate review

The review queue only lists candidates that pass a deterministic name filter and local-LLM authorship check (confidence ≥90% plus a verifiable quote). This is triage, not identity verification; a human still decides.

Candidate review requires prevalidation. LLM calls run only when requested.

Queue Reviewer

Candidate decision queue

Loading candidate reviews…

Reviewer identity is recorded for audit purposes. The current app does not authenticate reviewer identities; use this queue only in a trusted environment until authentication is enabled.

⚡ Total Obligations

Detected actionable items

⏳ Pending Obligations

Awaiting payment / signature

⚠️ Overdue Deadlines

Past target due dates

🎯 Auto-Struck (Cross-Matched)

ALCOA+ verified fulfillments

Operational To-Dos Ledger

Done	Task / Obligation	Type & Priority	Due Date	Amount	Source Document	Fulfillment / Cross-Match	Action
Loading operational obligations...

Sovereign Document Intelligence Service

The Process Scroller: 5-Stage Digitalization & Provenance Pipeline

Dual-Node Realtime Ollama Process Inspector (`/api/ps`)

Version Lineage & Document Evolution Explorer

Real-Time Telemetry & Diagnostic Event Stream

Multi-Signal GraphRAG Sovereign Explorer

3D Knowledge Graph Universe

Entity Name

GraphRAG-Powered AI Interrogation Assistant

3D Author-Candidate–Document Graph

Author candidate review

Candidate decision queue

The Process Scroller: 5-Stage Digitalization & Provenance Pipeline

Dual-Node Realtime Ollama Process Inspector (`/api/ps`)

Version Lineage & Document Evolution Explorer

Real-Time Telemetry & Diagnostic Event Stream

Multi-Signal GraphRAG Sovereign Explorer

3D Knowledge Graph Universe

Entity Name

GraphRAG-Powered AI Interrogation Assistant

3D Author-Candidate–Document Graph

Candidate decision queue

Entity Dossier & Risk Dynamics

Topical Cluster Entities