---

About

Founder of Datano — a data engineering & data science consultancy delivering ETL/ELT pipelines, data warehouses, ML solutions, custom dashboards, and API integrations. 50+ projects delivered across Azure (Data Factory, Databricks, Synapse), AWS, Python, SQL, and Power BI.

My work spans five main domains: AI/LLM agents & MCP tooling, data engineering & ML pipelines, cloud/DevOps & VPS orchestration, Microsoft 365 / SharePoint / Graph API integrations, and network automation (MikroTik, RouterOS, proxies).

Creator of pyaccelerate — a Python acceleration engine with multi-GPU, NPU, IoT, and Kubernetes support — and Animus, a self-hosted AI memory synchronization network.

Currently working on GPU-aware optimization for ollama and contributing to the AI inference ecosystem.

Tech Stack

---

Focus Areas

Domain	What I build	Key tech
AI / LLM / Agents	Memory sync networks, agent squads, MCP toolkits, local LLM optimization, transcription & diarization	Python, FastAPI, MCP, Ollama, Whisper, SQLite FTS5
Data Engineering & ML	ETL/ELT pipelines, GPU-accelerated analytics, NPS/verbatim classification, data warehouses, visual workflow orchestration	Python, Java, SQL, PL/SQL, CUDA, scikit-learn, KNIME, Pandas
Cloud, DevOps & VPS	Multi-project VPS orchestration, Docker stacks, CI/CD, licensing systems, boot/recovery tools	Docker, Nginx, PostgreSQL, Redis, Linux, PowerShell
Microsoft 365 & SharePoint	Graph API integrations, Power Automate MCP, SharePoint CRUD, mail archiving	TypeScript, PowerShell, Microsoft Graph, SharePoint
Network & Infrastructure	MikroTik/RouterOS automation, load balancers, proxies, hotspot managers, network optimizers	RouterOS Script, Python, Shell, C
Mobile & Embedded	Android toolkit, device rooting, multiboot USB, cross-platform legacy browsers	Python, Kotlin, Swift, Java, JavaScript
ERP & Web	Integrated ERPs, portals, landing pages, Markdown-to-HTML pipelines	Python, TypeScript, JavaScript, HTML/CSS

---

Public Projects

Project	Description	Tech
pyaccelerate	High-performance Python acceleration engine -- CPU, multi-GPU, NPU, IoT, K8s, auto-tuning	Python, CUDA, Cython, Rust
ollama-accelerate	Automatic hardware detection and performance optimization for Ollama	Python
python-gpu-statistical-analysis	GPU-accelerated statistical analysis with CuPy/CUDA and NumPy fallback	Python, CUDA
win-inventory-portable	Portable Windows inventory toolkit for collecting and exporting system information	PowerShell
bf6-optimizer	Battlefield 6 performance & network optimization scripts for Windows	PowerShell
CSVToolBox	Toolkit for processing CSV, Excel and tabular formats	Python
CSVSplitter	Split large CSV files preserving charset, separator and formatting	Python
Loader	CLI tool for loading CSV data into Oracle databases	Java
Spooler	Efficient Oracle Database spool execution and CSV report generation	Java
adb-toolkit	Complete Android backup, recovery & transfer with GPU acceleration	Python
IEPortable	Cross-platform legacy IE browser with Wine ActiveX for DVRs/cameras	JavaScript
PyExeBuilder	Simple compiler for Python scripts	Python

---

Private Activity

Beyond public repos, I maintain 48 private projects spanning:

Language	Projects
Python	17
PowerShell	7
TypeScript	5
Shell	4
HTML	3
Java	2
Other	2
JavaScript	2
C#	2
RouterOS Script	2
PHP	1
Jupyter Notebook	1

Areas include data engineering, AI/LLM tooling and MCP servers, visual ETL workflow orchestration, web apps and ERPs, network automation (MikroTik/RouterOS, proxies, load balancing), Microsoft 365/SharePoint integrations, VPS orchestration, and general systems tooling.

---

pyaccelerate

Metric	Value
Version	v0.11.0
Modules	27
Test Files	19
Stars	0
GPU Vendors	NVIDIA, AMD, Intel, ARM
Platforms	Linux, Windows, macOS, Android/Termux, IoT/SBC

---

AI & ML Projects

Animus — AI Memory Synchronization Network

Animus is a private, self-hosted memory synchronization network for AI assistants. It collects, indexes, curates, and syncs context across machines through a secured hub:

• Multi-assistant collection: GitHub Copilot Chat and CLI, Claude Code, Antigravity, project instructions, prompts, agent definitions, and curated notes.
• Knowledge retrieval: SQLite FTS5 search across sessions, projects, decisions, and reusable technical context.
• Knowledge OS: Verified material is promoted into a structured Obsidian vault, keeping raw memory separate from curated knowledge.
• Secure synchronization: Signed HTTPS requests, replay protection, peer identity and version visibility through a self-hosted hub.
• Peer lifecycle: Automatic release checks plus idempotent ZIP packages for installation, update, or repair on Linux and Windows.
• MCP and Web: MCP retrieval for compatible assistants and a sanitized Markdown viewer with Mermaid diagram rendering.

The MCP interface makes the network practical day-to-day: assistants can retrieve relevant notes, prior decisions, project conventions, and previous sessions on demand without exposing infrastructure details publicly.

Tech: Python, FastAPI, Pydantic, SQLite FTS5, HTTPS, HMAC-SHA256, Obsidian, Mermaid, Docker, systemd, Windows Task Scheduler, MCP.

Multi-Agent Orchestration Platform

Private platform for coordinating specialist AI squads through reusable workflows, explicit task routing, contextual memory, and Builder/Validator quality gates. Publicly shareable patterns include multi-stage execution, specialist handoffs, and evidence-based validation; agent definitions, infrastructure, and operational data remain private.

NPS Classification Engine

Enterprise feedback classification system for NPS verbatims. It runs an auditable cascade that separates customer feedback from operational boilerplate and classifies each record by product, reason, sub-reason, sentiment, root cause, and recommended action.

Layer	Technique	Role
0	Separator	Extracts customer feedback vs. operator boilerplate
1	Deterministic rules	Priority regex -- fast and auditable
2	Custom ML model	TF-IDF (word + char) + ensemble (LogReg + ComplementNB)
3	Pluggable AI	Low-confidence escalation to LLM (opt-in)
--	Fallback	Manual review when confidence is too low

Key design decisions:

• Confidence measured as margin between top-2 classes: top1 / (top1 + top2).
• Multi-scenario architecture: core/ engine + profiles/ plug-ins; the first profile is verbatim NPS, new domains add new profiles without touching core.
• Status bands: CLASSIFICADO (≥ 0.90), PENDENTE (0.40–0.90), ANÁLISE_MANUAL (< 0.40).

Tech: Python, scikit-learn, pandas, regex-based rules, pluggable LLM layer.

Repository is private and corporate-bound, so only technical approach and public patterns are shared here.

IRobot-Not

Proprietary desktop ETL workflow orchestrator built from scratch with Java 23 + Swing. It provides a visual, graphical interface for designing, scheduling, and executing data pipelines without relying on enterprise ETL suites.

• Visual pipeline designer: drag-and-drop style workflow composition via Swing UI.
• Multi-source connectivity: JDBC integration with Oracle, SQL Server, PostgreSQL, MySQL, and flat files.
• Execution engine: multi-threaded orchestration with ExecutorService, retry logic, error handling, and execution logging.
• Extensible architecture: plugin-friendly transformation steps and configurable pipeline definitions.
• Observability: real-time execution status, logs, and failure diagnostics directly in the UI.

Tech: Java 23, Swing, JDBC, ExecutorService, JSON/YAML, Maven, Oracle, SQL Server, PostgreSQL, MySQL.

Repository is private and proprietary; only the technical approach and public architecture patterns are shared here.

Selected Private Systems (Anonymized)

System	Public architecture summary
Integrated operations ERP	Modular commercial, field-service, project, inventory, finance, HR, accounting, supplier, fleet, and contract workflows
Franchise operations platform	Web portals for operational workflows, feedback, authentication, and structured business data
Secure network operations portal	Controlled remote network management with encrypted connectivity, auditability, and role-based access
Microsoft 365 integration gateway	Application-authenticated automation for mail, files, and collaboration services with least-privilege access
Accessible music technology	Assistive score interpretation combining optical recognition, structured music notation, audio, and accessible interaction

Client identities, domains, hostnames, repositories, environment topology, credentials, and business-specific rules are intentionally omitted.

---

Open Source Contributions

Forks & Branches

Fork	Upstream	Branch	Status
phpmixbill	ibnux/phpmixbill	GuilhermeP96-patch-1, GuilhermeP96-patch-2	--
ADModule	samratashok/ADModule	feature/ad-query-tool-v2	PR Open
ollama	ollama/ollama	api, bmizerany/client-registry, bmizerany/embedspeedup	PR Open
Aios	harshitgavita-07/Aios	feat/sdk-streaming-hardware-aware	PR Merged

Pull Requests

PR	Repository	Status	Date
feat: Ollama SDK integration, streaming UI, hardware-aware model selection	harshitgavita-07/Aios		2026-03-27
feat: autotune - automatic hardware-based performance optimization	ollama/ollama		2026-03-26
feat: Add comprehensive AD user query tool v2.0	samratashok/ADModule		2026-01-06
Added Brazilian Portuguese language.	GuilhermeP96/phpmixbill		2017-04-03
Brazilian Portuguese language added	GuilhermeP96/phpmixbill		2017-04-03

---

GitHub Stats

---