Project

Quantitative Algorithmic Trading & Reinforcement Learning Lab (PoC)

Ktrade Securities

Quantitative development environment and reinforcement learning algorithmic trading engine developed at Ktrade Securities, powered by vectorbt, TimescaleDB, and Redis.

PrototypeCompany · Ktrade SecuritiesClosed source1 Mar 2023 — 31 Oct 2023 · 7 mo

vectorbtPythonPostgreSQLTimescaleDBRedisNumPyPandasReinforcement LearningAlgorithmic TradingQuantitative Development

A quantitative finance and algorithmic trading research laboratory engineered at Ktrade Securities to investigate high-throughput vectorized backtesting and reinforcement learning decision policies for financial markets:

Vectorized Strategy Backtesting: Built using vectorbt and Numba, enabling rapid matrix-based backtesting, parameter optimization sweeps, and drawdown analysis across multi-year asset histories without iterative loop bottlenecks.
High-Frequency Time-Series Storage: Built on PostgreSQL with TimescaleDB hypertables, automating chunk partitioning, tick compression, and continuous aggregates for low-latency candlestick generation.
In-Memory Quote & Order Book Cache: Integrated Redis for high-speed simulated order matching queues, real-time spread tracking, and live quote caching.
Reinforcement Learning Execution Engine: Structured state-action-reward simulation environments using NumPy and Pandas to evaluate policy convergence for automated execution timing and slippage reduction.

Problem

Iterative backtesting frameworks are prohibitively slow for exploring complex parameter spaces, and standard relational engines struggle to ingest and compress high-frequency financial tick data.

Approach

Architected a vectorized research and simulation pipeline using vectorbt, TimescaleDB, PostgreSQL, Redis, NumPy, and Pandas, training RL policy agents on historical tick and candlestick feeds.

Results

Demonstrated sub-second multi-parameter optimization sweeps across tick datasets, automated continuous OHLCV rollups with TimescaleDB, and verified RL agent convergence under simulated transaction costs.

Connected work

Skills

Projects

Tracker Failure Simulator — Visual Multi-Object Tracking & Evaluation Lab

Experience

Senior Software Developer at Ktrade Securities

Loading content…

Vectorized Strategy Backtesting: Built using vectorbt and Numba, enabling rapid matrix-based backtesting, parameter optimization sweeps, and drawdown analysis across multi-year asset histories without iterative loop bottlenecks.
High-Frequency Time-Series Storage: Built on PostgreSQL with TimescaleDB hypertables, automating chunk partitioning, tick compression, and continuous aggregates for low-latency candlestick generation.
In-Memory Quote & Order Book Cache: Integrated Redis for high-speed simulated order matching queues, real-time spread tracking, and live quote caching.
Reinforcement Learning Execution Engine: Structured state-action-reward simulation environments using NumPy and Pandas to evaluate policy convergence for automated execution timing and slippage reduction.