AI Platform & Infrastructure
The shared serving, model-routing, and ML infrastructure that powers AI across the company.
Articles in AI Platform & Infrastructure
tsindex: Giving Coding Agents a Code Index Instead of a File Dump
Why we built a tree-sitter code index for coding agents, what real workloads taught us about tool design, and where our benchmarks show gains in speed, tokens, and cost.
Esteban Torres · · 16 min read
tsindex on DeepSWE: Less Context, Lower Cost
A paired DeepSWE task run with GPT-5.5 shows how structural code navigation can reduce time, tokens, and cost—and what a single comparison can tell us.
Esteban Torres · · 7 min read
From Model Sprawl to a Shared ML Platform
How we consolidated dozens of one-off model deployments into a single FastAPI-based, Kubernetes-served paved path that carries everything from classic prediction services to LLM agents and shared MCP tool servers — without turning shared infrastructure into a bottleneck.
LegalZoom Engineering · · 7 min read
Putting Machine Learning in the Checkout Path Without Making Checkout Depend on It
How we run multiple production ML predictors inside a latency-sensitive purchase funnel — and why the funnel never blocks on inference.
LegalZoom Engineering · · 8 min read
One API in Front of Every LLM: Building a Cost-Aware Model Gateway
Cost-per-token is a sticker price that tells you almost nothing. After an autonomous agent quietly burned through its monthly model budget and went dark for two days, we built a gateway that optimizes for what actually matters: realized cost per completed task.
LegalZoom Engineering · · 8 min read