Build. Break.
Document. Repeat.

Technical deep-dives, architecture decision records, and engineering war stories from the teams building HarchOS, SENSE, and Africa's sovereign compute platform. Written by engineers, for engineers.

6 Articles
Updated Weekly
InfrastructureAdvancedMarch 202624 min read

Inside HarchOS: How We Built a Distributed AI Operating System from Scratch

A complete walkthrough of HarchOS internals — from the custom scheduler and GPU topology awareness to the SENSE/THINK/ACT pipeline that orchestrates 1,798 GPUs across three data centers. No off-the-shelf orchestrator could handle our requirements, so we built one.

Read Technical Deep-Dive

Engineering Archive

BackendAdvanced19 min read

Designing the SENSE Layer: Real-Time Data Ingestion at 10M Events/Second

How we built a Kafka-free ingestion pipeline using a custom Rust-based event router that handles 10M events/sec with p99 latency under 4ms. Includes our benchmarking methodology and failure mode analysis.

AI/MLAdvanced16 min read

Our GPU Scheduling Algorithm: Balancing Throughput and Fairness Across 1,798 GPUs

We rejected bin-packing and FIFO for a weighted fair queuing approach with topology-aware placement. This post covers the math, the trade-offs, and why latency-sensitive inference workloads changed everything.

SecurityIntermediate14 min read

Zero-Trust Networking in Multi-Tenant AI Infrastructure

When every customer runs arbitrary CUDA kernels on shared hardware, network isolation is existential. We describe our SPIFFE-based identity layer, eBPF firewall, and runtime threat detection.

DevOpsIntermediate12 min read

From Terraform to Production: Our Infrastructure-as-Code Journey

Managing 200+ resources across Morocco, Senegal, and Cote d'Ivoire required more than Terraform modules. We built a custom provider, a drift detection system, and a deployment pipeline that validates before it applies.

InfrastructureAdvanced15 min read

Latency Optimization: How We Achieved Sub-12ms Inference for African Markets

Most LLM inference benchmarks assume US-East to US-East. We optimized for Casablanca-to-Dakar, Tunis-to-Lagos, and achieved p95 inference under 12ms using speculative decoding, model quantization, and edge caching.

Join the Engineering Team

We're hiring distributed systems engineers, ML infrastructure specialists, and security engineers across Casablanca, Dakar, and remote. Build systems that matter.

View Open Roles

Engineering RSS Feed

Subscribe to the engineering blog RSS feed. New posts, architecture decision records, and post-mortems — delivered straight to your reader.

Copy Feed URL