Technical Publications
Research, applied engineering and technical publications
from Agentic AI Systems, exploring locally hosted artificial
intelligence, enterprise integration and intelligent automation.
PUBLICATION 202610-AASPUB-0001 | 19 OCTOBER 2026
Enterprise AI Compute Node: Local Qwen AI Integration
with Microsoft 365 and the Future of Intelligent
Business Automation
This technical publication presents the demonstrated integration
of a locally hosted Qwen language model with Microsoft 365
email services through an independently operated AI Compute Node.
It also explores future enterprise capabilities including
financial document OCR, invoice and quotation extraction,
finance-ready CSV generation, voice-controlled assistance,
intelligent email drafting, calendar organisation and
human-authorised timesheet approval workflows.
Author: Reddy (Eddie) Jamisola
Organisation: Agentic AI Systems
Document Number: 202610-AASPUB-0001
View / Download Publication (PDF)
TECHNICAL PUBLICATION • 202610-AASPUB-0002 • 19 OCTOBER 2026
DF Memory — Three-Tier Persistent AI Memory
Proven and demonstrated DF Memory engineering by Agentic AI Systems,
featuring persistent RAG, TOC.md knowledge navigation, ontology,
evidence-based knowledge promotion and DF Smart Memory
AI-agent-coordinated retrieval.
A centralised memory architecture enables authorised distributed
AI nodes to retrieve and reuse organisational knowledge independently
of individual language models.
Author: Reddy (Eddie) Jamisola
Organisation: Agentic AI Systems
Document Number: 202610-AASPUB-0002
Read / Download Publication 0002 (PDF)
TECHNICAL PUBLICATION • 202610-AASPUB-0003 • 20 OCTOBER 2026
Dynamic GPU Compute Mode Switching, Model Switching and Multi-Inference
Proven single-GPU and multi-GPU AI execution, concurrent
multi-model inference and dynamic compute resource allocation,
developed by Agentic AI Systems.
The publication presents demonstrated configurations including
single-GPU multi-inference, multiple models on one GPU,
multi-GPU model execution and flexible allocation of five GPUs
between shared-model and independent-model workloads.
Author: Reddy (Eddie) Jamisola
Organisation: Agentic AI Systems
Document Number: 202610-AASPUB-0003
Read / Download Publication 0003 (PDF)
TECHNICAL PUBLICATION • 202610-AASPUB-0004 • 20 OCTOBER 2026
Consumer GPU Power-Curve Optimisation and Active Thermal Management
Experimental observations of 70–75% GPU power limits
with approximately 1–5% performance degradation across
selected NVIDIA and AMD consumer GPUs.
The research examines NVIDIA GeForce RTX 3080 and RTX 3090,
and AMD Radeon RX 6800 graphics cards, including silicon-lottery
variability, 100% fan operation during compute workloads
and a 30-second post-compute cooling period.
Results may vary between GPU models and individual chips,
even within the same model family.
Author: Reddy (Eddie) Jamisola
Organisation: Agentic AI Systems
Document Number: 202610-AASPUB-0004
Read / Download Publication 0004 (PDF)
TECHNICAL PUBLICATION • 202610-AASPUB-0005 • 20 OCTOBER 2026
DF Nexus: Unified Agentic AI Workflows Across Edge,
Home Lab, Distributed GPU and Cloud Infrastructure
DF Nexus is the foundational distributed agentic AI
architecture developed by Agentic AI Systems, connecting
smartphones, laptops, CPU-based edge devices, home laboratory
GPU infrastructure, distributed AI nodes and cloud computing
into one coordinated execution environment.
The architecture supports CPU-only, CPU+iGPU and CPU+dGPU
processing, small 1B and 4B language models, speech-to-text,
text-to-speech, single-GPU compute-mode switching,
model switching, multi-inference and dynamic multi-GPU
resource allocation.
WireGuard and Tailscale provide private encrypted
connectivity between authorised AI nodes, while cloud VPS
services coordinate workflows, persistent memory and
resource selection. Elastic GPU compute and authorised
frontier-model APIs extend available processing capacity.
Author: Reddy (Eddie) Jamisola
Organisation: Agentic AI Systems
Document ID: 202610-AASPUB-0005
View Technical Publication (PDF)
TECHNICAL PUBLICATION • 202610-AASPUB-0006 • 21 OCTOBER 2026 • UPDATED 14 NOVEMBER 2026 (U001)
DF Codex V1.1 — Engineering Persistent AI Knowledge
Through Vector-Based Semantic Retrieval
DF Codex is an incremental AI-assisted source-code analysis
system developed by Agentic AI Systems, incorporating Smart Archive,
repository snapshots, SHA256 evidence verification, incremental
processing, historical checkpoints and recovery mechanisms.
This living technical publication documents integration with
DF Memory Ontology Edition and the development of vector-based
semantic retrieval using local ONNX inference and
384-dimensional embeddings.
November 2026 Engineering Update (U001):
- Controlled update register and revision history.
- Architecture schematics and software stack documentation.
- Verified V1 operational capabilities and audit findings.
- Hierarchical synthesis implementation status.
- Component and worktree reduction limitations.
- Future development proposals and acceptance criteria.
The publication distinguishes operationally accepted functionality,
implemented but unverified capabilities, incomplete components
and future engineering proposals.
Author: Reddy (Eddie) Jamisola
Organisation: Agentic AI Systems
Document ID: 202610-AASPUB-0006
Latest Update: 14 November 2026
Revision: U001
Status: Living Technical Publication — Active Development
View Technical Publication (PDF)
TECHNICAL PUBLICATION • 202610-AASPUB-0007 • 22 OCTOBER 2026
AIGENESIS — Five-GPU AI Inference Architecture,
Dynamic Model Switching and Multi-Inference
A comprehensive engineering publication examining the
AIGENESIS five-GPU AI compute platform, including
motherboard, CPU, system memory, GPU hardware,
PCIe topology, model deployment and inference configurations.
The paper documents nine GPU and model execution modes,
concurrent inference benchmarks, multi-agent workload
orchestration, model switching, CPU/RAM offloading,
performance observations, system limitations and
experimental validation results.
Author: Reddy (Eddie) Jamisola
Organisation: Agentic AI Systems
Document ID: 202610-AASPUB-0007
View Technical Publication (PDF)
TECHNICAL PUBLICATION • 202610-AASPUB-0008 • 27 OCTOBER 2026
AI Intelligence Benchmark V1.0
A comparative evaluation of 24 local AI model configurations
across STRIX HALO and AIGENESIS. The publication examines
reasoning, systems engineering, cybersecurity, mathematics,
inference throughput and model accuracy, presenting benchmark
rankings, performance measurements, detailed scoring,
historical comparisons and technical findings.
Author: Reddy (Eddie) Jamisola
Organisation: Agentic AI Systems
Document ID: 202610-AASPUB-0008
View Technical Publication (PDF)
TECHNICAL PUBLICATION • 202610-AASPUB-0009 • 24 OCTOBER 2026
AIGENESIS — PCIe x1 vs x16: Does It Matter for Local AI Inference?
This study examines whether PCIe x1 connectivity is practical for concurrent, GPU-resident local AI inference, compared with an x16-connected GPU. Five independent llama.cpp servers were pinned to NVIDIA GPUs by CUDA UUID. Each model received three concurrent LONG requests of 512 generated tokens per GPU. Three rounds of simultaneous server startup-to-ready measurements were also recorded for each model.
Author: Reddy (Eddie) Jamisola
Organisation: Agentic AI Systems
Document ID: 202610-AASPUB-0009
View Technical Publication (PDF)
TECHNICAL PUBLICATION • 202611-AASPUB-0010 • 1 NOVEMBER 2026
NVIDIA Nemotron RE-USE — AI-Powered Background Music Removal and Vocal Isolation
Agentic AI Systems demonstrates NVIDIA Nemotron RE-USE running locally
on dual NVIDIA GeForce RTX 3080 hardware. This technical research
evaluates instrumental background music suppression and vocal
preservation using CUDA-accelerated AI inference, with FLAC and MP3
audio outputs. The publication documents the experimental methodology,
verified software environment, results and research limitations.
Author: Reddy (Eddie) Jamisola
Organisation: Agentic AI Systems
Document ID: 202611-AASPUB-0010
View Technical Publication (PDF)
Watch Demonstration Video
TECHNICAL PUBLICATION • 202611-AASPUB-0011 • 13 NOVEMBER 2026
EDYBOY BADBOY — Retro Arcade Beat-'Em-Up
Local AI-Assisted Game Engineering on STRIX HALO.
Agentic AI Systems documents the development of a retro pixel-art
side-scrolling beat-'em-up using Phaser 3, Vite and locally hosted
AI assistance involving Qwen 3.8 Flash Next. The research covers
sprite animation integration, JavaScript build verification,
development server testing and engineering limitations.
Author: Reddy (Eddie) Jamisola
Organisation: Agentic AI Systems
Document ID: 202611-AASPUB-0011
View Technical Publication (PDF)
TECHNICAL PUBLICATION • 202611-AASPUB-0012 • 14 NOVEMBER 2026
DF COMMANDER — Edge-to-Server Voice Email Assistant
TUF laptop speech recognition, Bonsai 1.7B, AIGENESIS Qwen3-30B-A3B,
Microsoft 365 email integration and Piper text-to-speech.
A technical engineering publication documenting distributed AI voice assistance.
Author: Reddy (Eddie) Jamisola
Organisation: Agentic AI Systems
Document ID: 202611-AASPUB-0012
View Technical Publication (PDF)
Watch Demonstration Video