I turn mathematical ideas into working software. Trading research, learning machines, thoughtful algorithms—and the engineering that brings them to life.
Explore the question, the implementation, and what the evidence taught me.
A Freqtrade strategy tested with 15-minute signals, 1-minute execution detail, historical funding, and 0.07% costs per side—plus look-ahead and recursive-stability audits.
A bid/ask-aware Nasdaq 100 and S&P 500 event engine built around US macro releases. The work mapped how attractive headline results collapsed under realistic cost and leverage constraints.
A walk-forward A-share framework with point-in-time announcements, Rank IC/ICIR, quantile portfolios, turnover costs, and explicit audits for survivorship and label leakage.
PPO trading research using dual LSTM encoders, supervised pretraining, real P&L tracking, state-aware action masking, and rolling validation across multiple model generations.
React + FastAPI system for news aggregation, sentiment, event calendars, and market forecasting. Holdout tests showed the first forecasting models failed to beat naive baselines.
An agent pipeline that reads architecture documents and diagrams, produces structured specifications, and generates a runnable educational game with traceable artifacts and tests.
An MQL5 event strategy with session-aware entry/exit, gap-risk sizing, and a detailed execution audit. Profitable in available history, but constrained by incomplete broker data.
A read-only real-time collector that detects sub-second sweep activity and liquidity exhaustion, then exports normalized events into the strategy research pipeline.
A scoped research direction for detecting correctness and optimization regressions across quantum compilers using equivalent-circuit transformations and scalable testing oracles.
A privacy-first concept for inferring routines from permissioned device signals, estimating confidence, and dynamically replanning a day without reading private content.
A causal XAUUSD research stack spanning signal construction, Dukascopy M1 execution data, spread and commission modeling, timeframe comparison, stop-loss recovery, and trailing-exit studies.
An MQL5 execution system combining liquidity sweeps, BOS/CHOCH, session logic, multi-timeframe structure, synchronized daily-loss controls, and ONNX model inference.
A hierarchical candlestick-model pipeline that loads and unloads 1H/4H CNN components on demand, integrating news sentiment before later migration toward reinforcement learning.
A four-person university project where I led software modeling across use cases, domain models, system sequence diagrams, operation contracts, and iterative architecture delivery.
Selected implementations connect causal inputs, stateful execution, transaction costs, and explicit rejection criteria. They support the finance profile; they do not define it.
Market candles, bid/ask ticks, funding, earnings timestamps, financial announcements, and news are aligned to the information actually available at decision time.
Inspect the validation design, cost assumptions, risk profile, and final conclusion behind selected systematic-finance experiments.
Passed the stated frequency and CAGR gates after modeled fees, slippage, funding, and bias checks. Still requires dry-run validation; profit factor was 1.129 and worst trade was −18.17%.
Charts use reported aggregate backtest metrics. No equity path is reconstructed or invented where time-series data is unavailable.
Signals, labels, earnings estimates, and announcements are frozen at the information actually available when a decision is made.
Fees, spread, slippage, funding, quote gaps, non-atomic fills, and missing history belong in the experiment—not in a footnote.
Backtest archives, exact commands, audit reports, and versioned implementations turn a result into an engineering artifact.
What I used, where I used it, and the engineering problem it solved.
Software engineering, intelligent systems, mathematical research, and rigorous experimentation developed through professional work and independent projects.
AI-enabled quantitative-finance research: data pipelines, factor construction, AutoML, financial-text signals, model validation, risk analysis, and reporting.
REST APIs, authentication, database query work, caching, logging, monitoring, and API documentation in a Java service environment.
Software development, testing, and debugging across Python, JavaScript, C++, and Java in an agile team.
Computer science foundations spanning algorithms, data structures, probability, statistics, linear algebra, software engineering, and computer architecture.
Two paper projects in progress. Proposed methods and next steps are separated from completed implementations and validated findings.
How can agent-generated software be verified beyond a successful demo?
Can equivalent circuits expose semantic errors and optimization regressions across compiler toolchains?