Voice-based daily logging for home daycare providers - Part 2

Voice-based daily logging for home daycare providers - Part 1

Design doc — Background, Assumptions, High-Level Design

Feature Store

One sentence: A feature store is a shared library of pre-computed inputs that machine learning models need to make predictions — so every team uses the same numbers, and those numbers are always up to date.

Design a Distributed Training System

Data parallelism, tensor parallelism, pipeline parallelism, gradient synchronization, fault tolerance — training a 70B model is a distributed systems problem as much as an ML problem.

The Dashboard was green while the model was lying

Open AI’s GPT-4o sycophancy rollback - and what it teaches us about why standard SRE metrics are nearly useless for LLM Systems

When the AI Deleted Production: The Kiro Incident

How Amazon's AI coding assistant caused a 13-hour outage — and what it teaches us about safe AI agent infrastructure design.

Fraud Detection at Scale

Rule engines, gradient boosting, graph neural networks — and why the hardest part isn't detection, it's winning an arms race.

Search Ranking System

Candidate retrieval, feature engineering,learning to rank

Design an AI Agent Infrastructure

Tool calling, memory, orchestration, fault tolerance — the infrastructure layer that turns LLM calls into reliable autonomous systems.

Design a RAG Pipeline at Scale

Chunking, Embedding, Hybrid Retrieval, Reranking

Model Serving Platform

Autoscaling, A/B Routing, Shadow Traffic, SLO Guarantees