CORE-CUT: Field Tested (Vixion 150cc)

tradeLLM: Live Cent Test (USC)

ChatLLM: v1.0 Android APK

🌸 Vercel Free Tier

🌸 ギャップ萌え // GAP MOE CERTIFIED

✨ Personal Workbench & Systems Log // ALTER EGO MODE

[ 仮想空間・研究室 ]

Krishna Alter Ego 🌸

Explorations across embedded & automotive systems and quantitative trading logic.

Focusing on systems operating directly in real-world environments with rigorous protection: from engine ignition timing control to protect the physical engine, to automated algorithmic trading with strict bounded risk. Built with a failsafe-first methodology and AI collaboration guided by explicit technical specifications.

View Projects ↓ Philosophy & Approach → Workbench & Lab 🛠️

🌸 Mascot Squad Guide // 仲間たち

Secret Pastel Lab

Institutional-grade engineering retained 1:1, accompanied by 4 interactive anime companions!

⚡ ツンデレ 🔪 ヤンデレ 🎈 元気 🌸 感謝

✨ 01 // Philosophy & Approach

How & Why I Build

Insights on how I design systems, respect physical constraints, and collaborate with AI.

Built on Curiosity & Practical Needs

I am an enthusiast exploring technology in my spare time without formal credentials in electrical engineering or computer science. Every project here starts from technical curiosity and personal needs—built to be deployed on my daily bike and personal market analysis.

#curiosity-driven #field-tested

Empirical & Observable in the Field

For me, a project is only complete once proven stable under real-world conditions. From inductive coil discharge voltage on a physical engine to live tick execution latency on broker terminals. Simulations are useful, but the ultimate benchmark is the field.

#empirical-testing #failsafe-first

Human Logic, AI Implementation

I leverage AI to implement modular C++ and Python code. However, the physical rules, circuit behavior, and non-negotiable safety guardrails (such as Anti-Missfire and 16-Gate Risk filters) are strictly defined beforehand through disciplined specifications before any code is generated.

#specification-driven #zero-hallucination

⚡ CORE PRINCIPLE: HUMAN DEFINES THE SPECIFICATION, AI IMPLEMENTS THE CODE

AI excels at structuring algorithms rapidly, but cannot feel engine heat, mechanical shift dynamics, or bear financial risk. Understanding physical constraints—when ignition can safely be interrupted, IRAM isolation on the ESP32, and maximum daily drawdown—remains strictly driven by human specifications (AGENTS.md).

🌸 Mascot Squad Consensus: Human specification leads, AI turns it into robust code!

🌸 02 // Featured Systems

Projects & Experiments

Three engineering domains unified by the same standard: deterministic precision, transparency, and failsafe execution.

FIELD TESTED Embedded & Automotive Yamaha Vixion 150cc ⚡ バカ! 点火制御 // 200ns

CORE-CUT v1.0 ESP32

DIY ESP32 Ignition Cut Module & Real-Time Telemetry for EFI Motorcycles

Core-Cut [ツンデレ]

Tsundere ⚡

"I-It's not like I engineered this 200ns anti-missfire cutoff because I was worried about you! I just don't want my circuitry wrecked by an amateur, baka!"

✦ click for anime voice line

💢

Standalone ignition control module based on the ESP32, engineered and road-tested on a daily Yamaha New Vixion Advance 150cc. Implements racing features like Quickshifter, Pit Limiter, Idle Rumble, and Launch Control while safeguarding engine integrity via the Anti-Missfire protocol.

▸ Anti-Missfire: Current interruption strictly on Rising Edge; never cut during Dwell charging

▸ Power Path: High-side P-MOS IRF5210 (+12V) with automatic relay failsafe bypass

▸ FreeRTOS Dual-Core: Core 1 dedicated to zero-delay engine timing; Core 0 for web & telemetry

▸ Dual-Trace Scope: Real-time visual verification ensuring cut pulses never collide with dwell

ESP32 DevKit V1 FreeRTOS IRAM-Assembly PlatformIO C++ Modular P-MOS IRF5210 WebSocket 10Hz LittleFS

Key Telemetry Spec

Cut Latency: < 200 ns (Inline)

Clock Target: 240 MHz Dual-Core

Overheat Lock: > 115°C Hard-Cut

Web Dashboard: 16 Monochrome Menus

Live Web Demo ↗ GitHub Repository ↗

ANTI-MISSFIRE DUAL TRACE

[Digital Cycle Visualizer]

Sample: Last 16 Engine Cycles

■ DWELL WINDOW ■ CUT — COIL OFF ■ CUT HELD THROUGH DWELL

Cut asserted as dwell began (race condition): 0

Cycles where cut was held through dwell: 0

No cut overlapped a dwell window in the last 16 cycles. Engine firing normally.

LIVE CENT TEST (USC) Quant & Financial AI MetaTrader 5 Native 🔪 愛の16ゲート // ZERO LOSS OBSESSION

tradeLLM Hybrid Quant + AI

Institutional-Grade Algorithmic Forex Trading & Multi-Agent Intelligence

TradeLLM [ヤンデレ]

Yandere 🔪❤️

"You are not allowed to lose even a single pip, understand? If you ever try to bypass my 16 gates... I'll lock your broker terminal forever so you'll only rely on me~ ❤️"

✦ click for anime voice line

🔪❤️

Automated forex trading system with strict separation of concerns: Python handles 100% deterministic mathematical calculations (technical indicators, support/resistance, ATR), while the LLM provides macro reasoning and directional context to eliminate numeric hallucination.

▸ 16-Gate Risk Engine: 16 sequential risk validation gates before orders reach the broker

▸ DataBridge TCP: Low-latency local socket bridge connecting Wine Linux MT5 and Python

▸ Fresh Tick Re-Fetch: Re-polls live broker quotes post-AI inference to prevent price drift

▸ Circuit Breaker: Enforced daily drawdown ceiling, max 3 concurrent trades, and Friday flat close

Python 3.11 MetaTrader 5 Native Wine Linux IPC Statistical Arbitrage Tavily & Perplexity Circuit Breaker Post-SL Cooldown 16-Gate Engine

Risk & Pipeline Spec

Architecture: 9-Agent Pipeline

Risk Filter: 16 Gates (Strict)

Max Open: 3 Trades Global

Weekend Flat: Auto-Close (20:00 UTC)

Architecture Spec ↗ GitHub Repository ↗

16-Gate Hard Risk Engine Simulator [Sequential Risk Gate Evaluation]

Pipeline: AnalysisAgent → RiskAgent (16 Gates) → MT5

SELECT TEST SCENARIO:

EVALUATION OUTCOME:

STATUS: APPROVED (ORDER DISPATCHED)

All 16 risk gates passed. Order approved and dispatched to MT5 broker terminal.

ACTIVE DEV Mobile & AI Systems Flutter / Android Native 🎈 もぐもぐ // OFFLINE RAG POWER

ChatLLM v1.0 APK

Developer AI Chatbot for Android (Terminal Aesthetic)

ChatLLM [元気娘]

Playful 🎈✨

"Yahoo~! What are we exploring today? I've already digested your 500-page document into SQLite WAL, hehehe! ✨"

✦ click for anime voice line

✨

A high-performance Android AI chatbot application built with Flutter, designed with a disciplined terminal-inspired interface (Aider / Claude Code aesthetic). Engineered with client-side autonomous tool execution (Tavily & Exa AI RAG), zero token pruning with manual /compact memory checkpoints, and an AES-256-GCM encrypted local vault.

▸ Autonomous Tool Loop: Client-side agentic search (Tavily/Exa) with 0ms pre-fetch streaming

▸ Zero Token Pruning: 100% full context memory retention without silent server summarization

▸ BYOK Multi-Model: Native Anthropic, OpenAI, DeepSeek, Ollama, and custom API proxies

▸ Vault & Local Storage: SQLite WAL + FTS5 full-text search with AES-256-GCM encryption

Flutter 3.x Dart Android Native SQLite WAL / FTS5 Tavily & Exa RAG AES-256-GCM LaTeX Engine BYOK Multi-Model

App & Engine Spec

Platform: Android (Flutter/Dart)

Context Policy: Zero-Pruning + /compact

Local Storage: SQLite WAL + FTS5

Vault Security: AES-256-GCM Local

Android APK Release ↗ GitHub Repository ↗

Android Terminal UI Preview [Aider / Claude Code Flat Aesthetic]

UX: Android APK • Pastel Messenger CLI

SELECT DEMO SCENARIO:

GENERAL ASSISTANT web:tavily

ctx: 3.1k/128k • 2% cache: 74% ▾

Tokens: 520 in (cache: 384 • 74%) • 140 out • 660 total

↺ retry ⎘ copy

🛠️ 03 // Environment & Gear

The Workbench (/rig)

Hardware, lab instrumentation, and development stack used during experimentation.

🔌 Hardware & Physical Lab

▪
Test Bike: Yamaha New Vixion Advance 150cc (EFI)
▪
MCU: ESP32 DevKit V1 (Xtensa LX6 240MHz Dual-Core)
▪
Sensors: Hall Effect NJK-5002C, NTC 50k, Optocoupler PC817
▪
Power Circuit: P-MOS IRF5210, BJT 2N2222, SPDT Relay Songle
▪
Lab Gear: Digital Multimeter, custom etched PCBs, logic probe

📈 Trading Pipeline & Data Stack

▪
Execution Terminal: MetaTrader 5 on Wine Linux
▪
Bridge: EA DataBridge v3.0 via local TCP Socket (:9090)
▪
Quant Engine: Python 3.11+ / 3.13, Library ta, Pandas
▪
Database: SQLite 3 (Signals, Trade Journal, Audit Logs)
▪

💻 Cloud & Development

▪
OS / Host: Debian GNU/Linux (Workstation & Cloud Server)
▪
▪
▪
▪

🎵 04 // Soundtrack

Workbench Rotation

Curated music played while assembling hardware, inspecting signals, or monitoring markets.

LAB CASSETTE DECK 🌸

PLAYING • 320 KBPS

SIDE A • C-90 HI-BIAS TYPE II • CHROME STEREO REC

WEST COAST CLASSIC

Money Trees

Kendrick Lamar (ft. Jay Rock)

Tape Counter: 04:20

🔥 CURATED LAB ROTATION

8 Tracks

🎮 05 // GitHub Activity

Live Contribution Snake Runner

Commit activity from @Little-Krishnaa on GitHub (last 15 weeks), visualized with an autonomous pathfinding snake runner.

LIVE SYNC Window: Last 15 Weeks (105 Days)

Commit Target: 11

GitHub Profile ↗

🍎 COMMITS CONSUMED

0 / 11

🐍 SNAKE LENGTH

4 segments

⭐ ACTIVITY SCORE

0 pts

🎯 SIMULATOR STATUS

Hunting Commits...

Less More

💡 The snake navigates autonomously using pathfinding to reach the nearest unconsumed commit without colliding with its own body. 100% Client-Side Rendered