I built and operated a portfolio of independently deployed production systems, standardizing observability while delivering coordinated improvements to reliability, performance, and resource efficiency. For multi-asset, multi-strategy interday equity options, I built a Kubernetes backtesting framework with indexed job scheduling and an autonomous gpt-oss-120b-backed signal-mining agent that identified strategies with Sharpe above 1.5.
I also fine-tuned Amazon Chronos-T5 with LoRA on labels derived from the backtesting engine, then used the resulting embeddings to train per-asset TabPFN classifiers for inference.
I consolidated multiple TradeStation market-data collectors into a reusable Terraform module and calibrated a jump-, rate-, and dividend-aware stochastic-volatility model to live option-chain snapshots. I implemented Bayesian anomaly detection over model-backed forecasts and fed narratives of traded-expiry implied-volatility smile repricing to gpt-oss-120b for trade decisions.
The deployed systems achieved 5–10% weekly returns on capital at risk. An automated daily evaluation harness measured reliability, rule compliance, consistency, and robustness.
I deployed a QLoRA-tuned Llama-3.1-8B and DSPy prompt-optimizer-backed FX trading system. Its news-relevance pipeline combined BM25, Qwen3 Embedding, reciprocal rank fusion, and Qwen3 Reranker; gpt-oss-20b compressed news and price action into the inference context. Evaluation-gated GEPA prompt optimization used gpt-oss-120b as the reflection model.
I fine-tuned a frozen NF4 Llama-3.1-8B base with rank-16 LoRA using single-node multi-GPU DDP, then evaluated held-out reward, accuracy, and precision for base and adapted policies using vLLM continuous batching. An OpenAI-compatible vLLM endpoint provided versioned adapter promotion and storage-backed rollback.
I built a daily rebalanced, regime-aware equity portfolio using a Hidden Markov Model for regime detection, ElasticNet forecasts, and NSGA-II optimization, achieving 5% annualized alpha in a multi-year backtest. I also shipped a multimodal RAG PDF assistant backed by Qwen3-VL, a two-pass workflow, and Qdrant.
