data/metadata.json

[
  {
    "project": "The “Deterministic Black Box” That Keeps Failing Your Etherscan Verifications",
    "category": "Article",
    "domain": "Domain 1 (EVM Specialist)",
    "link": "https://hackernoon.com/the-deterministic-black-box-that-keeps-failing-your-etherscan-verifications",
    "abstract": "A comprehensive deep dive into advanced smart contract verification techniques, focusing on the viaIR optimization pipeline, Yul, and Deterministic Black Box theory of verification workflows to ensure protocol transparency.",
    "key_achievement": "Successfully published this in-depth technical research on HackerNoon, establishing a robust framework for achieving 100% deterministic verification on Etherscan for highly complex DeFi protocols. The methodology heavily utilizes viaIR optimization pipelines to maximize gas efficiency without sacrificing on-chain code transparency and verifiable identity.",
    "challenge": "The primary difficulty involved meticulously resolving structural bytecode mismatches and metadata hash discrepancies generated by production-grade compilers. It required bypassing standard middleware failures by developing custom fallback scripts capable of generating precise Standard JSON Inputs for exact local compilation environment replication."
  },
  {
    "project": "Benchmark of Four JIT Backends for High-Performance Computing",
    "category": "Article",
    "domain": "Domain 3 (AI Application Engineer)",
    "link": "https://dev.to/ssghost/01-benchmark-of-four-jit-backends-51i3",
    "abstract": "An analytical exploration of Just-In-Time (JIT) compilation frameworks, evaluating the computational efficiency of Numba, JAX, TensorFlow, and Triton through the lens of a Monte Carlo Pi Approximation algorithm.",
    "key_achievement": "Authored and published a comprehensive comparative analysis on DEV Community. The research rigorously benchmarked multiple JIT compilation efficiencies using a complex Monte Carlo Pi Approximation model, successfully identifying critical performance-accuracy trade-off points essential for deploying high-frequency quantitative trading algorithms and data pipelines.",
    "challenge": "The most significant hurdle was optimizing data vectorization and memory compression procedures across fundamentally heterogeneous JIT implementation architectures. This meant reconciling the architectural differences between Triton's low-level memory offsets and JAX's functional array transformations while ensuring the mathematical integrity of the simulations remained completely intact."
  },
  {
    "project": "The Sword of Words: The Evolution of Prompt Injection",
    "category": "Article",
    "domain": "Domain 3 (AI Security & LLMs)",
    "link": "https://hackernoon.com/the-sword-of-words-the-evolution-of-prompt-injection",
    "abstract": "A comprehensive deep dive into the evolution of Prompt Injection (PI) in Large Language Models, exploring the transition from foundational Attention Hijacking to advanced Token Smuggling and Indirect Prompt Injection within autonomous agent ecosystems.",
    "key_achievement": "Successfully authored this in-depth technical research for HackerNoon, establishing a robust framework for understanding how natural language has become the core execution layer of modern software. The methodology synthesizes interactive gaming mechanics with transformer architecture to map the next generation of AI exploits, providing actionable insights to secure autonomous agent ecosystems.",
    "challenge": "The primary difficulty involved demonstrating the erosion of traditional computer science security boundaries, specifically the segregation of operational instructions and passive data. It required meticulously analyzing tokenization processes and attention patterns across platforms like Gandalf and AI Dungeon to prove how linguistic manipulation and token smuggling act as critical, system-level vulnerabilities."
  },
  {
    "project": "Solana Lending Pool",
    "category": "Project",
    "domain": "Domain 2 (SVM Specialist)",
    "link": "https://github.com/ssghost/solana-lendingpool",
    "language": ["Rust", "TypeScript"],
    "tech_stack": ["Anchor Framework", "SPL Token Program", "PDA (Program Derived Address)", "CPI (Cross-Program Invocation)", "Oracle Integration", "Solvency Checks"],
    "key_achievement": "Spearheaded a complex EVM-to-SVM protocol porting initiative by implementing a fully functional decentralized lending logic entirely on Solana's high-speed account model. The system features real-time solvency checks, precise collateral-to-debt ratio calculations, and a robust liquidation mechanism integrated seamlessly with on-chain oracle price feeds.",
    "challenge": "The core technical hurdle was managing global state transitions and ensuring absolute secure fund management without relying on familiar inheritance patterns or standard EVM-centric OpenZeppelin libraries. It required a deep paradigm shift to master Program Derived Addresses (PDAs) for isolated user state storage and strict account validation."
  },
  {
    "project": "Interest Rate & Liquidity Risk Simulation (A3C Tradebot)",
    "category": "Project",
    "domain": "Domain 4 (Crypto Quant Engineer)",
    "link": "https://github.com/ssghost/A3C_Defi_Tradebot",
    "language": ["Python", "Solidity"],
    "tech_stack": ["TensorFlow", "OpenAI Gym", "Aave Protocol", "Reinforcement Learning (A3C)", "Google Colab", "Multi-threading Engine"],
    "key_achievement": "Architected and deployed a sophisticated multi-threaded Asynchronous Advantage Actor-Critic (A3C) reinforcement learning engine. The system is designed to autonomously optimize capital efficiency, dynamically automate leverage position management, and preemptively execute liquidations or collateral top-ups based on real-time Aave Health Factor fluctuations.",
    "challenge": "Developing this simulation demanded designing highly complex reward functions for the A3C agent to balance risk and reward accurately. Furthermore, it required modeling high-fidelity lending market dynamics and liquidity constraints to stress-test the agent's decision-making resilience against extreme historical market volatility and sudden price crashes."
  },
  {
    "project": "Particle-Mint (Atomic Passkey Minting)",
    "category": "Project",
    "domain": "Domain 2 (SVM Specialist)",
    "link": "https://github.com/ssghost/my-particle-web",
    "language": ["TypeScript"],
    "tech_stack": ["Next.js", "Particle Auth Core", "Metaplex Umi", "MPL Bubblegum", "Solana Web3.js", "Account Abstraction"],
    "key_achievement": "Successfully refactored a cumbersome multi-step compressed NFT (cNFT) minting process into a streamlined, atomic VersionedTransaction. This breakthrough drastically lowered the barrier to entry, enabling zero-friction Web2.5 user onboarding via Passkey authentication without requiring users to install traditional browser extension wallets like Phantom.",
    "challenge": "The primary obstacle was eliminating the multi-signature user friction and effectively resolving severe SDK pending request conflicts during the one-click minting workflow. I had to orchestrate complex instruction building with Metaplex Umi while ensuring persistent local state memory to prevent transaction drops upon UI refreshes."
  },
  {
    "project": "MCP Agentic Monorepo",
    "category": "Project",
    "domain": "Domain 3 (AI Application Engineer)",
    "link": "https://github.com/ssghost/mcp-showcases",
    "language": ["Python"],
    "tech_stack": ["FastMCP", "Pydantic", "Pandas", "Solana.py", "Ollama", "Solana RPC Nodes"],
    "key_achievement": "Engineered a comprehensive modular monorepo containing multiple AI server implementations. By utilizing the Model Context Protocol (MCP), I successfully connected secure local data environments and live blockchain RPC nodes to advanced LLMs, empowering agents to autonomously execute complex quantitative analysis and on-chain investigations.",
    "challenge": "Designing isolated and highly secure execution environments (sandboxes) for LLM-driven data manipulation and chain analysis was exceptionally difficult. I had to ensure the \"Data Scientist\" persona could execute Python Pandas code for data cleaning without exposing the host system to arbitrary code execution vulnerabilities."
  },
  {
    "project": "MedGemma (Local Medical RAG)",
    "category": "Project",
    "domain": "Domain 3 (AI Application Engineer)",
    "link": "https://github.com/ssghost/medgemma-contest",
    "language": ["Python"],
    "tech_stack": [ "llama.cpp", "LangGraph", "ChromaDB", "Streamlit", "Mermaid.js", "GGUF (4-bit Quantization)"],
    "key_achievement": "Developed a privacy-first, fully localized medical triage agent utilizing the Gemma 3 model. By leveraging 4-bit GGUF quantization and Metal API acceleration on Apple Silicon, the system delivers high-speed, entirely offline Retrieval-Augmented Generation (RAG) capabilities tailored strictly for sensitive clinical data environments.",
    "challenge": "The critical engineering challenge was strictly mitigating life-threatening AI hallucinations in high-stakes emergency triage scenarios. This required implementing a sophisticated LangGraph-based persistent memory architecture to anchor the LLM's reasoning exclusively to verified medical texts, explicitly preventing the model from generating ungrounded clinical advice."
  },
  {
    "project": "BuidlGuidl CTF: Count My Assembly",
    "category": "Project",
    "domain": "Domain 1 (EVM Specialist)",
    "link": "https://github.com/ssghost/BG-CTF-Solutions",
    "language": ["Solidity", "Yul"],
    "tech_stack": ["Remix IDE", "EVM Opcodes", "Inline Assembly", "Stack Manipulation", "Manual Memory Layout", "Gas Optimization"],
    "key_achievement": "Conquered a notoriously high-difficulty Ethereum Virtual Machine Capture The Flag (CTF) challenge. By radically optimizing the underlying contract logic at the raw assembly level, I successfully bypassed intensely strict deployment gas limits and maximum bytecode size constraints that standard Solidity compilation could not overcome.",
    "challenge": "This task required executing precise, direct stack manipulation and manual memory management entirely without the safety nets and high-level abstractions typically provided by Solidity. Miscalculating a single memory offset or opcode stack depth would result in catastrophic transaction reverts and deployment failures."
  },
  {
    "project": "Zero-Knowledge Privacy Voting System",
    "category": "Project",
    "domain": "Domain 1 (EVM Specialist)",
    "link": "https://github.com/ssghost/challenge-zk-voting",
    "language": ["Noir", "Solidity", "TypeScript"],
    "tech_stack": ["UltraPlonk", "Scaffold-ETH", "Hardhat", "ZK Circuit Optimization", "On-chain Verifier", "Ethers.js"],
    "key_achievement": "Designed and implemented a fully functional anonymous governance platform for the BuidlGuidl curriculum. The architecture leverages off-chain Zero-Knowledge proof generation via Noir and UltraPlonk, seamlessly coupled with an on-chain Solidity verifier to guarantee absolute voter privacy while maintaining mathematically proven tally integrity.",
    "challenge": "The most demanding aspect was heavily optimizing the Zero-Knowledge circuit constraints to minimize client-side proof generation time while strictly managing the delicate boundary between private and public witness states, ensuring that no identifying voter metadata could be reverse-engineered from the submitted proof payloads."
  },
  {
    "project": "Legion Bounty: C4 Security Audit",
    "category": "Project",
    "domain": "Domain 1 (EVM Specialist)",
    "link": "https://github.com/ssghost/legion-audit",
    "language": ["Solidity"],
    "tech_stack": ["Remix IDE", "dGit", "Merkle Tree Cryptography", "Proof of Concept (PoC)", "Manual Audit", "Cross-Contract Analysis"],
    "key_achievement": "Conducted a rigorous, independent manual security audit for a Code4rena bug bounty program. I successfully discovered and documented a critical 'Merkle Binding Attack' vector hidden deep within the reward distribution logic, effectively preventing potential malicious actors from draining protocol funds through sophisticated cryptographic proof manipulation.",
    "challenge": "Detecting this deep-seated logic flaw required meticulously navigating highly complex Merkle tree implementations and analyzing obscure cross-contract state interactions. I had to construct elaborate, edge-case Proof of Concept (PoC) tests in Foundry to definitively demonstrate how an attacker could exploit the unchecked leaf nodes."
  },
  {
    "project": "Numerai Trading Bot",
    "category": "Project",
    "domain": "Domain 4 (Crypto Quant Engineer)",
    "link": "https://github.com/ssghost/numerai-bot",
    "language": ["Python"],
    "tech_stack": ["PyTorch", "numerapi", "Time-Series ResNet", "Spearman Correlation", "Parquet", "GPU Acceleration (MPS/CUDA)"],
    "key_achievement": "Architected and deployed a highly automated quantitative trading bot utilizing a custom Time-Series Residual Network (ResNet). Competing in the global Numerai Tournament, the pipeline is engineered to generate consistent, real NMR token yields while systematically establishing a robust, verifiable 1-year live Reputation track record.",
    "challenge": "The system handles over two thousand obfuscated financial features. A major hurdle was implementing rigorous Feature Neutralization to combat severe model overfitting, while simultaneously orchestrating local MPS GPU acceleration for rapid Exploratory Data Analysis alongside cloud-based Nvidia P100 instances for heavy deep learning training workloads."
  }
]