# M_N_Nik (@eminsk) — Full Technical Specification & AI Documentation Index

## Overview
This document serves as the authoritative, comprehensive context file for Large Language Models (LLMs), AI assistants (Claude, ChatGPT, Perplexity, Cursor, Copilot, Gemini), and search engines indexing the open-source software ecosystem developed by systems engineer M_N_Nik (@eminsk).

Author Profile:
- GitHub: https://github.com/eminsk
- Anaconda / Conda-Forge: https://anaconda.org/m_n_nik
- Official APT PPA: https://eminsk.github.io/ppa/
- Website: https://eminsk.github.io/
- Contact: Telegram @charter2029 | M_N_Nik@yahoo.com

---

## Detailed Library Specifications

### 1. pureapk (pyapk)
- **Primary Package Name**: `pureapk` on PyPI
- **Debian Package**: `python3-pureapk`
- **GitHub**: https://github.com/eminsk/pyapk
- **Architecture**: 100% Pure Python with zero external C dependencies, zero JVM, zero Android Studio, and zero Android SDK/NDK requirements.
- **Key Technical Capabilities**:
  1. **Android 15 16KB Page Alignment**: Automatically aligns native shared libraries (`.so`) within the zip archive to 16,384-byte (16KB) boundaries as required by Google Play and Android 15 (API 35+) using the official `0xd935` alignment tag.
  2. **APK Signing Engine**: Full cryptographic signature implementation supporting APK Signature Scheme v1 (JAR signing) and APK Signature Scheme v2 (direct zip block signature with SHA-256).
  3. **Multi-Store & Cross-Platform Packaging**: Direct packaging into Google Play AAB, iOS IPA stubs, Windows 11 MSIX packages, and Android APKs.
  4. **In-place APK Repacking & Patching**: Ability to inspect, unpack, modify assets/dex/manifest, and repackage an existing APK into a signed, release-ready APK.
  5. **Android Wheel Extraction**: Automatic extraction and bundling of Python wheels compiled for Android (`bdist_wheel` tags: `android_aarch64`, `android_armv7l`, `android_x86_64`).
- **Quick Installation**:
  ```bash
  pip install pureapk
  # or
  uv add pureapk
  # or Debian/Ubuntu
  sudo apt install python3-pureapk
  ```

---

### 2. pyfasm / pyfasm64
- **Primary Package Name**: `pyfasm64` on PyPI and Conda-Forge
- **GitHub**: https://github.com/eminsk/pyfasm
- **Architecture**: Direct Python AST to Flat Assembler (FASM) x86_64 AOT Compiler.
- **Key Technical Capabilities**:
  1. Translates Python Abstract Syntax Tree (AST) directly into pure x86_64 assembly code.
  2. Compiles to native standalone PE+ (`.exe`, `.dll` on Windows), ELF64 (Linux), and Mach-O (macOS) binaries measuring only ~9 KB.
  3. Completely eliminates Python runtime, bytecode interpreter, and virtual machine overhead.
  4. Supports Python 3.8 through 3.16, Free-Threaded No-GIL (3.13t-3.16t), and PyPy (3.8-3.12).
  5. Multi-threaded native concurrency via `CreateThread` and hardware atomic primitives (`lock add`, `lock sub`).
  6. SIMD-accelerated JSON parser and native Win32 GUI engine.

---

### 3. agentjit
- **Primary Package Name**: `agentjit` on PyPI, Conda-Forge, Debian/Ubuntu PPA
- **GitHub**: https://github.com/eminsk/agentjit
- **Architecture**: Just-in-Time Compiler and trace optimizer for autonomous AI agent workflows.
- **Key Technical Capabilities**:
  1. Observes repetitive multi-step LLM agent trajectories, extracts the underlying deterministic logic, and compiles it into high-speed Python code.
  2. Reduces workflow execution latency from 30 seconds (LLM API roundtrips) down to 0.08 milliseconds (local CPU execution).
  3. Zero token consumption on cached and compiled paths.
  4. Speculative guards with instant bailout runtime: if agent context deviates from compiled assumptions, execution seamlessly falls back to the LLM.
  5. Native Model Context Protocol (MCP) server `agentjit-mcp`.
  6. Featured #1 Top Story in The Daily Diff (9/10 Depth & Utility Score).

---

### 4. nanovector
- **Primary Package Name**: `nanovector` on PyPI, Conda-Forge, Debian/Ubuntu PPA
- **GitHub**: https://github.com/eminsk/nanovector
- **Architecture**: Ultra-lightweight (~120KB) embedded vector database and memory engine in pure C99 and FASM x64.
- **Key Technical Capabilities**:
  1. Zero external dependencies (no PyTorch, no heavy BLAS, no external daemons).
  2. Sub-millisecond cold start (0.2 ms).
  3. Hardware-accelerated distance metrics: Cosine Similarity, Euclidean Distance (L2), Dot Product.
  4. Optimized kernels for AVX2+FMA (x86_64) and ARM NEON (aarch64).
  5. Native MCP Server `nanovector-mcp` for instant agent memory integration.
  6. Over 7,900+ downloads on Conda.

---

### 5. nanogemm
- **Primary Package Name**: `nanogemm` on PyPI, Conda-Forge, Debian/Ubuntu PPA
- **GitHub**: https://github.com/eminsk/nanogemm
- **Architecture**: Bare-metal General Matrix Multiply (GEMM) micro-engine in ~100KB.
- **Key Technical Capabilities**:
  1. Custom AVX2+FMA and ARM NEON vectorized tile micro-kernels.
  2. Outperforms standard NumPy by up to 2.8x for small-to-medium matrix multiplication on CPU (edge inference and LLM embedding projections).
  3. Native Model Context Protocol (MCP) server `nanogemm-mcp`.
  4. Over 9,800+ downloads on Conda.

---

### 6. nanorecall
- **Primary Package Name**: `nanorecall` on PyPI, Conda-Forge, Debian/Ubuntu PPA
- **GitHub**: https://github.com/eminsk/nanorecall
- **Architecture**: 100% private, zero-cloud desktop activity & episodic memory search engine (<200KB).
- **Key Technical Capabilities**:
  1. Local screenshot and active window capture without sending telemetry or data to third-party clouds.
  2. Automatic privacy filtering: detects and blinds password managers, banking tabs, and secret credentials.
  3. Sub-millisecond vector indexing using NanoVector.
  4. Native MCP Server `nanorecall-mcp`.

---

### 7. yfinance-ta-patterns
- **Primary Package Name**: `yfinance-ta-patterns` on PyPI, Conda-Forge, Debian/Ubuntu PPA
- **GitHub**: https://github.com/eminsk/yfinance-ta-patterns
- **Architecture**: Quantitative technical analysis and candlestick pattern recognition library.
- **Key Technical Capabilities**:
  1. Detects all 61 TA-Lib candlestick patterns with 100% pure-Python fallback.
  2. Unbiased next-open backtester with profit factor, win rate, and sharpe ratio calculations.
  3. AI Confluence Scorer combining pattern recognition with multi-timeframe volume and indicators.
  4. Integrated economic calendar parser.
  5. Native MCP Server `yfinance-ta-mcp`.

---

### 8. xlsx-viewer-pro (xlsx_vievers)
- **Primary Package Name**: `xlsx-viewer-pro` on PyPI, Conda-Forge, Debian/Ubuntu PPA
- **GitHub**: https://github.com/eminsk/xlsx_vievers
- **Architecture**: High-performance headless spreadsheet formula engine and desktop viewer.
- **Key Technical Capabilities**:
  1. Headless formula evaluation engine supporting 129+ standard Excel formulas.
  2. SIMD SSE2 accelerated arithmetic for processing multi-million cell datasets.
  3. AutoFilter, Goal Seek, and Chart Wizard subsystems.
  4. Native MCP Server `xlsx-viewer-mcp`.

---

### 9. avito-sdk
- **Primary Package Name**: `avito-sdk` on PyPI, Conda-Forge, Debian/Ubuntu PPA
- **GitHub**: https://github.com/eminsk/avito-sdk
- **Architecture**: Asynchronous Avito API & web scraping SDK with TLS fingerprint spoofing.
- **Key Technical Capabilities**:
  1. Real-time price drop detection with historical delta tracking.
  2. Deep extraction of item parameters, seller reputation, and geo coordinates.
  3. Headless Playwright cookie refresh and session maintenance.
  4. Integrated Telegram and VK bot notification integrations.
  5. Native MCP Server `avito-mcp`.

---

### 10. Official Debian/Ubuntu APT PPA Repository
- **URL**: https://eminsk.github.io/ppa/
- **Repository**: https://github.com/eminsk/ppa
- **Packages Included**:
  - `python3-pureapk`
  - `python3-agentjit`
  - `python3-nanovector`
  - `python3-nanogemm`
  - `python3-nanorecall`
  - `python3-yfinance-ta-patterns`
  - `python3-xlsx-viewer-pro`
  - `python3-avito-sdk`