Skip to content

SaaS · Applied AI

A grounded support assistant that scaled without a rewrite.

Sector
SaaS
Timeline
Multi-month build
Team
Senior, founder-led
Engagement
New build
No rewritescaled as traffic grew

The problem

Support volume was growing faster than headcount, and answers had to be correct and cite policy — a raw chatbot was a liability, not a help.

The engineering challenge

Answers grounded in the customer’s own documentation, measurably correct, with a guardrail and a human-in-the-loop path for anything low-confidence — all cheap enough to run at scale.

The solution

We built a retrieval pipeline over the customer’s knowledge base, an evaluation harness scoring every change against a labelled set, guardrails for scope and confidence, and a review queue for escalations. Capacity and cost were modelled up front so a large traffic increase needed no re-architecture.

Architecture

Technology

Chosen for this workload — the load, the team, and the ten-year cost.

TypeScriptPythonPostgreSQLVector DBsAnthropic

The result

Grounded
answers cite source policy
No rewrite
absorbed traffic growth
Faster
time-to-resolution

Start

Have a system like this?

Tell us the hard part — we'll come back with a technical point of view.

Or email [email protected]