E-Vision

Blog

Blog

Engineering notes by Tarek Fawaz on IoT telemetry, platform architecture, DevOps and AI

Before and after: a monolithic sequential RabbitMQ consumer versus three fault-isolated consumers with bounded channels, parallel workers and batch throttling
Before and after: a monolithic sequential RabbitMQ consumer versus three fault-isolated consumers with bounded channels, parallel workers and batch throttling
IoT Engineering

Re-Engineering a Telemetry Consumer for Resilience: How We Stopped Losing Data Under Load

How we split a monolithic RabbitMQ consumer into 3 fault-isolated services with bounded channels, parallel workers, and batch throttling, stopping data loss at 10,000 devices.

18 min readRead the article
Diagram of a fleet data collector under budget: 10,000 trucks feed a TCP receiver, LRU deduplication and key routing into RabbitMQ, guarded by a 512MB memory budget and a CPU budget under 5%.
IoT Engineering

Building an IoT Data Collector Service That Doesn't Eat Your Server: CPU & Memory Budgets for High-Throughput Fleet Telemetry

How to handle 10,000+ concurrent fleet connections under 512MB RAM, with LRU deduplication, config-driven routing, CPU budgets, and health-signal-driven stability.

15 min readRead the article
Diagram of a spec-driven multi-agent pipeline: specialized agents move a requirement from spec to plan to implementation, with human gates at each transition and full traceability.
AI & Agentic Workflows

How I Built a Multi-Agent Development Workflow That Writes, Plans, and Implements Specs Autonomously

A deep dive into building a spec-driven development system with 7 specialized AI agents, human-in-the-loop gates, and full traceability from requirements to code.

12 min readRead the article
Three frameworks before shipping an AI feature: data readiness assessment, build-vs-buy decision, and true cost model
AI Product Strategy

Before You Build AI Into Your Product: Data Readiness, Build vs. Buy, and the Cost Nobody Budgets For

Three frameworks every leader needs before shipping AI features: a data readiness assessment, a build-vs-buy decision tree, and a true cost model that exposes the categories most business cases miss.

14 min readRead the article
Before and after: an unbounded deduplication cache growing to 4.4GB versus a bounded LRU + TTL cache (50,000 entries, 5-minute TTL) at 35.9MB
IoT / Performance

How We Cut IoT Telemetry Memory Usage by 99%

A production IoT case study on reducing memory consumption in high-throughput TCP telemetry pipelines processing GPS and RFID data.

8 min readRead the article
DTCM architecture: a raw telemetry stream passes through a metadata layer, made up of Signal Registry, Dynamic Parser, Routing Engine (MQTT 5.0) and Configuration API, before reaching application logic, instead of signals hardcoded in firmware and server.
Architecture / IoT

Metadata-Driven Telemetry: Decoupling Signal Routing from Firmware

How metadata-driven telemetry reduces firmware coupling and improves operational agility for connected devices and embedded systems.

10 min readRead the article
Diagram: master, develop and feature branches map to stable and prerelease NuGet versions, computed in a Version stage and passed through Build to Publish
DevOps / .NET

Building a Production NuGet Pipeline with Azure DevOps

A practical DevOps article on automated NuGet package versioning, branch-based prereleases, and cross-stage variables in Azure DevOps.

6 min readRead the article
Two tracks side by side: building with AI as team enablement and building for AI as product features, joined by one principle: don't chase the trend, chase the value.
Leadership & AI Strategy

Building With AI vs. Building For AI: A Leadership Framework for Adoption Without Enforcement

A leadership framework for distinguishing team AI enablement from product AI features, and adopting both without mandates, friction, or wasted investment.

10 min readRead the article