Currently
Working at the intersection of formal hardware security verification and cognitive AI agent architectures. Currently engineering Blackbox OS for automated system testing.
Selected Research & Preprints
Neuro-Symbolic Hypothesis Engine: Autonomous Scientific Discovery via Hybrid Large Language Models and Symbolic Regression
An autonomous quantitative discovery agent that extracts governing mathematical laws from raw empirical datasets.
From Tool Selection to Autonomous Execution: Process Templates, Sandboxed Code, and Modular Orchestration for Reliable LLM Agents at Scale
Blackbox OS is a research prototype for reliable LLM tool use in quant/data-science settings: stage-local tool partitions, expert process templates (SOPs), sandboxed code execution, and light validation guardrails. Full multi-agent “desk” composition remains future work.
Asymmetric Hidden Markov Modeling of Order Flow Imbalances for Microstructure-Aware Market Regime Detection
Unlike traditional HMM applications focused on asset returns, this model incorporates directional asymmetry in transition dynamics and adjusts signal weighting based on local microstructure (spread-driven certainty).
Projects & Open Source
Blackbox-OS
A reliable, sandboxed execution runtime and modular orchestrator for LLM agents at scale with 8-stage causal analysis and AST loopback verification.
Neuro-Symbolic-Hypothesis-Engine
An autonomous quantitative discovery agent that extracts governing mathematical laws from raw empirical datasets.
asic-puzzle-2026-solution
End-to-end automated reverse-engineering flow extracting gate-level netlists from raw GDSII silicon layout geometry to solve the Jane Street ASIC challenge.
Open Source Upstream Contributions
NVIDIA CCCL (CUDA C++ Core Libraries)
Upstream CUDA parallel primitives and performance optimizations for GPU-accelerated C++ core algorithms.
Meta Velox Execution Engine
Hardware-aligned skew-resilient radix partitioner hook for high-throughput batch execution in Meta Velox.
Writings & Technical SOPs
High-Performance Zero-Copy Casting Kernels for Apache Arrow ListView Arrays
A deep-dive technical article detailing zero-allocation memory slicing, O(1) contiguous pointer paths, and 64-bit word bitmap scanning for Arrow compute expressions.
Reverse Engineering Jane Street's 2026 ASIC Puzzle: From GDSII Geometries to Symbolic Z3 Solving
A complete technical writeup detailing polygon extraction, footprint classification, diode bridge node recovery, and symbolic Z3 constraint modeling.
Contact
Open to research collaborations, formal security audits, and high-assurance systems engineering discussions.