Prominent Works.

The pipeline's hardest constraint wasn't throughput. It was the circuit-breaker mechanism. Analytic queries on derivatives data could take anywhere from 200ms to several minutes depending on the time window and complexity. When clients hit endpoints during peak market hours, a single slow query could cascade and starve faster ones.

We designed query slots with explicit timeouts; queries that exceeded their budget were terminated and fell back to a smaller pre-computed window. Clients always got a response, even if degraded. This single mechanism saved significant execution costs and was the difference between a system that worked under load and one that didn't.

The custom DataTable for CME Group was the second hard part. Real-time filtering and infinite scrolling over tens of thousands of analytics rows, previously only available via REST API. Memory optimization in React was non-trivial at this scale. We ended up with windowed virtualization and aggressive memoization.

The compliance automation was the highest-leverage piece. Engineers were spending most of a workday each week on cross-channel alert triage, log aggregation for audits, and routine compliance reports. We built integrated tooling that consumed DataDog signals and automatically generated the artifacts these tasks required.

The harder, less visible win was API latency. We took client-facing latency down by 55% through three concurrent moves: batched requests where independent calls were happening in tight loops, database partitioning to keep hot tables small, and selective denormalization to remove join hot paths. None of these were exotic. The difficulty was sequencing them without breaking production.

The hard part was the payments layer. Underneath a few-clicks experience sat a full neobank running on crypto rails: integrating payment processors and on-chain USDC settlement through Circle, tied together so that getting paid was a single link for the user while the system handled wallets, settlement, and payouts behind it. I designed the composable profile architecture that made it work, where a freelancer's public profile, their payment tiers, and their crypto payout flow were assembled from the same composable building blocks.

I led the build end to end. I built the platform solo, then hired and trained the first designer and the first developer and transitioned the codebase and architecture to the team that carried it forward. The product has since evolved into a global payments platform, and I moved on to other opportunities.

Verstack is an experimental proptech venture. As technical co-founder, I lead engineering end to end. Each inspection becomes a navigable spatial record rather than a folder of disconnected videos, photos, and notes. Property managers can understand what happened, where it happened, and which evidence supports the finding without reconstructing the inspection manually.

Syndic moves the agent workflow off the desktop. You dispatch a task from your phone, it runs CLI agent processes on a VM, and you monitor progress, review the diff, and merge the PR without opening a laptop. It is self-hosted and runs on your existing agent subscriptions rather than reselling tokens, so you keep your own plans and keys. I built the entire platform solo.

Provenant is not a logger, a database, a workflow engine, or a complete accounting product. It is the enforcement layer that sits between raw source bytes and the final answer a finance team is willing to release. It records where every output number came from and packages the evidence for verify and replay, enforcing four guarantees across every supported read path: non-bypassable exit, fail-closed errors, schema-first validation, and deterministic replay. It is built for SR 11-7 and FINRA 2026 contexts, where controllers, audit leads, platform engineers, and compliance officers need a verifiable audit chain from raw bytes through to a replayable audit bundle.

fbx produces a full set of balanced double-entry accounting files from a single seeded event stream: GL, bank statement, sub-ledgers, and financial reports, all of which reconcile against each other. The key property is determinism: the same seed, bundle, period, and version produce byte-identical output across runs, operating systems, and CPU architectures. That makes it usable as a stable fixture for testing financial pipelines, where you need data that is realistic, reconcilable, and reproducible.

BlockUnion was early full-stack blockchain infrastructure work. The premise was simple: if someone shared an experience with a decentralized service anywhere on the web, the platform could turn that contribution into a reputation signal and reward it with tokens that had monetary value.

I worked as a contract engineer on the first startup version and owned the architecture from scratch: choosing the technical stack, designing for elastic cloud capacity, connecting the application layer to Ethereum/Web3 flows, and translating business requirements into a production-ready technical system.

MTM was a different domain with the same core responsibility: turn an early startup idea into an implementable platform. As Principal Solutions Architect, I translated the business goal into a complete technical plan covering application boundaries, service endpoints, data distribution, cloud resources, and operational requirements.

The work was intentionally end to end. I defined the scalable infrastructure model, produced frontend and backend diagrams, planned logging, monitoring, alerting, security, testing, and DevOps, and selected the technology stack for each part of the platform.