| I'm Alex - a Senior Software Engineer who spent a decade in AI. I also create content through The AI Merge, sharing the lessons I've learned. |
Building and teaching production AI systems - through The AI Merge.
At Axon
- Built the OS-to-cloud stack for a fleet of thousands of edge devices (NVIDIA Jetson, Raspberry Pi, embedded Linux) — Wind River Linux images, RAUC for A/B OTA updates, and AWS IoT Core / Device Shadows for fleet configuration and telemetry.
- Shipped a fully on-edge DeepStream AI pipeline on Jetson — optimized models and an expert system, wired into the edge-AI service stack to stream results to the cloud. Also designed a dynamic tiered config system pushed to devices as heartbeats over IoT Device Shadows.
- Built AI pipelines analyzing video at scale — semantic video search, image captioning, video summarization, and synthetic data generation for training, served through Triton inference on edge devices and in Kubernetes.
- Helped build Search: ~25 microservices spanning multiple clouds, ingesting device data into a large-scale search and filtering layer that holds a low p99 under load.
- Built agentic systems for on-call and incident management — coding-agent workflows and agents that triage production traces.
- Pushes internally for better AI-assisted engineering: knowledge sharing, tutorials, and workshops on using AI well as an engineer.
| Kubrick — an MCP-based multimodal AI agent with eyes and ears: vision, voice, and memory in one open-source course, built with Miguel Otero Pedrido |  |
| MAVS — edge multi-agent vision system for wildlife conservation: an MLOps pipeline trains the CV models, then an MCP/A2A agentic layer runs them at the edge | Soon |
| Forge — a human-gated SDLC for building with AI agents: you grill the spec, design, and plan; agents execute inside that scope through implement, review, and ship | Soon |
| Patch — a local voice companion on an NVIDIA DGX Spark, with an iPhone as its mic and display: captures notes into Obsidian, runs Todoist, and handles daily focus/review routines | Soon |
I like to think about Software and AI grouped in three ladders, which is the way I also teach it.
Foundations
Systems
Engineering
We need less hype. More engineering. If you're a software engineer trying to actually understand AI, not just call an API, here's what's in it for you:
- The AI Merge — field notes and deep dives on production AI systems, no prompt lists. Grew from ~300 to 10,000+ subscribers in a year.
- Early access to The AI Atlas, a 220+ slide visual guide to the whole AI stack — hardware, training, inference, agents, deployment
- An NVIDIA content partnership — tutorials on CUDA, TensorRT, NIM, RTX, plus hardware access via an NVIDIA DGX Spark
- A YouTube channel with system-design walkthroughs and live coding, not tutorials that stop at "hello world"
 |
 |
 |
 |
