Skip to content

Portfolio

Ford Pro: EV Charging Platform

OCPP Gateway Migration Ford Pro

The Problem

The legacy monolith architecture could not scale to support a rapidly growing fleet of EV chargers, resulting in high infrastructure costs and maintenance bottlenecks. The challenge was to transition to a microservices architecture while ensuring seamless, uninterrupted OCPP communication for thousands of active chargers.

My Approach

I led the architectural strategy by authoring foundational RFCs to define the core platform services and the path for OCPP 2.x adoption. I prioritized a scalable gateway design capable of handling high-throughput real-time traffic while optimizing our observability stack to prevent future cost overruns.

What I Built

I architected and deployed a central communication gateway, successfully migrating 8,200 chargers to the new system. Alongside feature implementation, I entirely revamped our Datadog logging and monitoring configurations to filter logs more efficiently and accelerate root cause analysis.

Results

8,200 chargers migrated~2,150 servers decommissioned$9,000/mo cost savings60% Datadog cost reduction99.95% uptime

High-Throughput Kafka Fleet Consumer Ford Pro

The Problem

The platform required a robust mechanism to ingest and process a massive influx of daily telemetry messages from fleet chargers. Dropped messages or processing delays would directly impact the analytics and operational visibility for fleet managers.

My Approach

I focused on fault tolerance and throughput. Before writing code, I evaluated the partition strategy and consumer group configurations to ensure the system could scale horizontally as the fleet size and data volume grew.

What I Built

I architected and implemented a high-throughput Kafka consumer from the ground up. I built the service to reliably ingest, process, and route millions of messages daily, ensuring strict adherence to defined data contracts.

Results

6M+ messages/day99.9% data integrityHorizontal scalability

Global OCPP Simulator (GOS) Ford Pro

The Problem

Validating the OCPP Gateway at scale required a simulation platform that could realistically mimic charger behavior across thousands of devices. Without this, testing system performance and reliability before production rollout was not feasible.

What I Built

I drove the development of the Global OCPP Simulator, delivering a platform that simulates over 2,000 chargers to validate system performance and reliability at scale. I authored and drove the adoption of 5 foundational RFCs that defined the core architecture and feature roadmap for the platform.

Results

2,000+ chargers simulated5 RFCs authored

Ford Pro Charging (FPC) Platform Ford Pro

The Problem

The Ford Pro Charging SaaS platform needed a reliable backbone for all charger data systems, the services that fleet managers depend on for visibility into their EV charging infrastructure.

What I Built

Served as the Subject Matter Expert (SME) for all charger data systems, owning the backend services (NestJS, MongoDB) that form the backbone of the Ford Pro Charging platform for 8,200+ chargers.

Results

8,200+ chargers supportedSME for charger data

ChargeBox Simulator Ford Pro

The Problem

The existing charger simulation tool was slow and lacked automated test coverage, making it a bottleneck in the development and QA pipeline for charging session workflows.

What I Built

I overhauled the ChargeBox Simulator and implemented a test automation suite, significantly improving the speed and reliability of charger and charging session simulation.

Results

30% performance boost40% testing time reduction

Charging KPI Analytics System Ford Pro

What I Built

Developed a KPI analytics system to measure the effectiveness and ROI of Ford's managed charging algorithms, translating complex data into actionable business insights for stakeholders.

Jiralyzer: AI Analytics Platform Hackathon Finalist

The Problem

Analyzing massive volumes of Jira tickets for performance metrics was slow and manual. The goal was to build an intelligent, secure system that could query and summarize this data using natural language, without timing out on large datasets.

What I Built

I built a full-stack AI analytics platform (React, Flask, GPT-4) with enterprise-grade features including Azure AD authentication, automated PDF reporting, and a real-time AI chat for interactive performance analysis. Implemented intelligent caching and parallel processing for LLM workloads, cutting analysis response times by 70%.

Results

70% faster analysis100,000+ Jira issues processedHackathon Finalist

Experion Technologies: Enterprise Solutions

SCANCO / CloudConnectIT / SPA Experion Technologies

What I Built

Designed a full-stack solution (.NET, Angular, AWS) that integrated with Sage ERP to track over 2,000+ products through their entire manufacturing lifecycle. Contributed to a SaaS-based product for inventory tracking, increasing customer satisfaction by 25%.

Results

2,000+ products trackedSage ERP integration25% customer satisfaction increase

ARIA B2B / BAZAR (B2B2C) Experion Technologies

What I Built

Guided the frontend development team (React) to build and launch a B2B platform, scaling it to support over 5,000+ users and drive product sales.

Results

5,000+ usersReact frontend

WAREFLEX: Logistics Platform Experion Technologies

What I Built

Built a real-time logistics platform (.NET, Angular) to monitor 1,000+ active shipments and optimize warehouse scheduling with live dock availability data.

Results

1,000+ active shipmentsReal-time monitoring

Personal Projects

File-To-BinaryVideo-BackTo-File Golang

What I Built

Engineered an encoding mechanism that converts any file into a binary video format, enabling lossless decoding back to the original source file. An exploration of binary data representation, video encoding pipelines, and creative approaches to data storage.

View on GitHub →

PUB/SUB Implementation using Redis Golang

What I Built

A technical write-up and implementation of a highly available publish/subscribe messaging system using Redis. Covers connection management, message serialization, and reliable delivery patterns.

Read the Write-up →

PhotonicOps — Offline Telemetry Ingestion Engine Golang · gRPC · HIPAA

The Problem

Silicon photonic biosensors in HIPAA-sensitive clinical environments generate continuous, high-frequency optical resonance telemetry that must be ingested reliably with zero cloud dependencies — fully air-gapped, no OpenAI/AWS/GCP calls anywhere.

What I Built (Phase 0 + Phase 1 — implemented)

Designed and built a high-throughput, offline telemetry ingestion engine in Go:

  • High-throughput gRPC service (client-streaming RPCs) sustaining 10 kHz data ingestion, with a Protobuf-defined sensor telemetry contract and a synthetic load-generating client (mock 10 kHz sensor simulator with injected Gaussian noise + thermal drift modeling).
  • Zero-allocation / low-GC hot path using sync.Pool for buffer reuse; concurrent worker-pool pattern (fixed goroutine pool + buffered channel) for backpressure-aware load handling; lock-free / mutex-protected circular (ring) buffer for fixed-memory, leak-free long-running ingestion.
  • Go workspaces (go.work) for multi-module monorepo management.
  • GitHub Actions CI pipeline: build, go vet, race-detector test runs (go test -race), coverage reporting, lint gating, and cross-compilation to static Linux binaries as artifacts.
  • Fully offline Docker Compose infra stack (Postgres, Prometheus, Grafana, Langfuse, Ollama) pinned to linux/arm64 for Apple Silicon, with an environment-gate script validating service health before development proceeds.
  • Profiling instrumentation via net/http/pprof to verify GC-pause / performance SLAs.
  • ADRs documenting key tradeoffs: mTLS transport security, Unix-domain-socket IPC design, fail-safe LLM decision fallback states.

Architected, not yet built (later phases — designed via ADR)

  • A local-only agentic hardware-triage system on Ollama (self-hosted LLM, zero cloud API calls) with Langfuse tracing.
  • Safety-first agent pattern: triage agent must degrade to REQUIRES_MANUAL_REVIEW on timeout or low-confidence response — human-in-the-loop safety constraint for agentic actions in a clinical/HIPAA context.

AI-Assisted Engineering Practice

Used Claude Code as a primary agentic dev tool — configured project-level constraint files to enforce architectural rules (ARM64-only Docker configs, zero-allocation Go patterns, zero-cloud-API policy) automatically across every session. Designed custom directory-scoped AI personas per service, and used ADRs as durable machine-readable context for AI continuity.

Results

10,000 samples/sec sustainedZero-allocation hot pathRace-detector-cleanFully air-gapped