Portfolio

Projects

Representative work across distributed systems, robotics, and full-stack engineering.
Each project includes my specific role and measurable outcomes.

Distributed Systems

Scalable, fault-tolerant services powering multi-million record workloads with sub-100ms latencies.

High-Scale Pricing & Search API

150K+ daily queries • 15M+ records • <100ms P95

The Challenge

Searching 15M+ records is error-prone — users misspell names, use partial terms, and search across multiple fields. The API must serve both humans expecting instant responses and automated scripts making 100K+ calls with minimal total runtime.

The Solution

Distributed WebAPI serving 15M+ trading-card records with sub-100ms P95 latency via Elasticsearch.

Real-time CDC pipelines (Debezium/Kafka) stream mutations into Postgres + ES with staged imports and automatic rollback on validation errors.

Role

Lead architect & sole developer — system design, CDC pipelines, performance profiling, observability.

Tech Stack

Node.jsTypeScriptPostgreSQLKafkaElasticsearch
Impact
  • 150K+ queries served daily
  • 83% reduction in indexing latency
  • Zero timeout failures post-migration

Platform & Tooling

Developer infrastructure and operational tooling that accelerates team velocity and system reliability.

CDC Lifecycle Manager & Search Benchmarking Suite

Connector ops • Relevance evaluation • k6 load testing

The Challenge

Data pipeline failures are invisible until customers complain. Teams waste hours manually restarting connectors and have no objective way to measure if search improvements actually help users find what they need.

The Solution

TypeScript microservice for Debezium/Kafka connector lifecycle with health monitoring.

Paired with a FastAPI + SvelteKit benchmarking suite for side-by-side search relevance evaluation.

Role

Full ownership — design through production, CI/CD, monitoring.

Tech Stack

TypeScriptFastAPISvelteKitKafkak6
Impact
  • Incident response: hours → minutes
  • Data-driven relevance tuning enabled

Full-Stack Web

End-to-end web applications with complex RBAC workflows, real-time dashboards, and seamless integrations.

RBAC Ops Portal for CDC & Staging Pipelines

Mission-critical workflows • Real-time dashboards

The Challenge

Operations teams relied on Slack messages and shared spreadsheets to coordinate data imports — leading to duplicate work, missed approvals, and costly staging errors that slipped into production.

The Solution

Internal Next.js control panel replacing paper/Slack workflows with RBAC-secured approvals.

Handles CDC operations, image ingestion, job monitoring, and search debugging for production.

Role

Lead developer — UX design with operators, full-stack implementation.

Tech Stack

Next.jsReactRBACREST
Impact
  • 90% reduction in staging errors
  • 4+ hours/week manual work eliminated

Robotics & Control

High-throughput motion control systems with industrial-grade reliability and real-time telemetry.

180 items/min Industrial Sorting Robot

99.2% accuracy • 72-hour stress tested • Real-time telemetry

The Challenge

Manual sorting of high-value inventory is slow, error-prone, and doesn't scale. Human fatigue causes accuracy to drop over time, and there's no visibility into throughput or failure patterns.

The Solution

Real-time control stack for high-throughput industrial sorting robot.

Multithreaded Python/C++ motion control, industrial vision, barcode validation, and Prometheus telemetry.

Role

Systems architect — control flow, motion stack, sensor integration.

Tech Stack

PythonC++OpenCVModbusPrometheus
Impact
  • 99.2% accuracy at 180 items/min
  • 0 dropped connections over 72 hours

Robotics & Vision

Precision robotics integrating synchronized multi-camera capture with ML inference pipelines.

Multi-Camera ML Digitization System

Pneumatic pick-and-place • Synchronized capture • ML pricing

The Challenge

High-value collectibles require professional-grade photography for accurate pricing — but manual capture is too slow for scale, and inconsistent lighting causes ML models to misclassify condition and value.

The Solution

High-precision digitization: pneumatic vacuum handling, synchronized multi-camera capture, and ML inference for real-time pricing.

Built for high-value inventory requiring sub-millimeter mechanical tolerance.

Role

Full-stack robotics — mechanical concept through ML pipeline integration.

Tech Stack

PythonC++Multi-cameraML Pipelines
Impact
  • 10x faster hardware iteration
  • Production-ready for delicate inventory

Embedded & Hardware

Low-level systems from HDL design to firmware — custom RISC-V cores, FPGAs, and real-time sensing.

32-bit RISC-V Core & STM32 Firmware

Custom soft core • Transistor-level ALU • Real-time sensing

The Challenge

Understanding processor architecture requires hands-on implementation. Commercial cores are black boxes — building from transistors up reveals how instruction pipelines, ALU operations, and memory hierarchies actually work.

The Solution

Designed and fully simulated a 32-bit RISC-V soft core in VHDL with custom pipeline stages.

Built transistor-level Booth multiplier and STM32 firmware for environmental sensing/telemetry.

Role

Individual contributor — architecture, simulation, hardware testing.

Tech Stack

VHDLRISC-VSTM32FPGA
Impact
  • 32-bit RISC-V core fully simulated
  • Real-time STM32 telemetry firmware
  • Transistor-level ALU verification