Petaling Jaya · SS15

The week’s model calls, named by screen rather than buried in a total.

Model Vertex Point reconstructs how an AI-assisted app actually invokes its models — waiting composers, quiet save-hooks, retries, fallbacks — and writes that down so product and engineering are looking at the same rows.

Small working group talking over papers around a long table — Assessments are read with a counterpart who can name routes the same day — in our SS15 room or at your table in the Klang Valley.

01 · What we actually measure

Usage, as the app fires it

Call inventory: Every invocation site with a human name: composer, summariser, classifier, nightly rewrite. Retries sit beside the site, not in a separate fiction called “errors”.
Waiting versus not: Whether a person is looking at a spinner, or the app is talking to a model while nobody is in the room.
The gap list: Questions the logs still cannot answer. Inferred rows are labelled as inferred. We do not dress timestamp soup as certainty.

02 · Flagship work

Model usage assessment

A time-boxed reading of one production app. We take a call inventory, reconstruct a recent window of logs or traces, and hand back a written picture plus a ninety-minute walkthrough. Prompt rewriting and vendor shopping are out of scope; naming what actually fired is in.

Most assessments run ten to fifteen working days from confirmed access. Fees start at RM 8,400 and follow the number of invocation sites, not a subscription card.

Read the assessment

Notebooks and conversation at a meeting table

03 · Related work

When a full assessment is not the first conversation

Two colleagues looking at a laptop together during a review

Instrumentation review

A short, opinionated reading of whether your current logs can answer basic usage questions — before you commission a full assessment or change what you record.

Ask about a review

Two people in discussion across a table with notebooks

Stakeholder usage briefing

A facilitated reading of an existing usage picture — ours or yours — for a mixed room of product, operations, and engineering who do not share a vocabulary for model calls.

Arrange a briefing

Person working with documents at a bright desk near a window

Repeat usage snapshot

A second measurement of the same app after a release, a model change, or a logging fix — compared against the earlier picture rather than started from a blank page.

Request a snapshot

From a recent window

The chat composer accounted for the waiting completions. A background summariser that nobody had opened in nine days still fired on every record save.

That split — waiting versus not — is the sentence most rooms have not been able to say out loud. It is also the sentence a usage assessment is for.

04 · How an engagement starts

Access first, then the window

Name the app

One production app, a counterpart, and a week you can grant log access or a supervised sitting session.
Inventory the sites

We list invocation sites in language a product lead recognises, then mark waiting, retries, and fallbacks.
Read, write, walk through

The report comes before the conversation. The walkthrough is a reading, not a replacement for the document.

The method, step by step

05 · From the room

What people actually said afterward

They sat with our composer logs for a week and handed back a table our product lead could actually read. The nightly rewrite job — which none of us had put on the original list — was the row that changed the conversation.

Hafiz Rahman, Engineering manager, document assistant app, Subang Jaya — Model usage assessment

The first draft assumed we had request IDs on every line. We did not, and that cost two extra days while they rebuilt sequences from timestamps. I would have preferred they refused to start until we added the field. The route-by-route picture, once it arrived, was still the first one our ops lead did not argue with.

Mei Ling Ong, Head of product, internal help desk, Petaling Jaya — Model usage assessment

Journal

Open notebook with handwritten notes beside a laptop

2026-04-14

What belongs in a model call inventory

A call inventory is not a log dump. It is a named list of every place an app asks a model for something, written so a product lead and an engineer can point at the same row.

Person typing on a laptop in a bright room

2026-05-06

When people say the model is slow

Latency complaints in an AI-assisted app often describe a route, a retry loop, or a screen that waits on work it did not need to wait on — not a model that suddenly forgot how to type.

Printed documents and a fountain pen on a desk

2026-06-11

User-facing completions and the quiet ones behind them

A week of model usage is often two stories: the completions a person asked for, and the completions the app decided to run while nobody was looking.

Bring one named app

Tell us roughly how many screens invoke a model and a week you can grant access. We will say whether an assessment can even start.

Request an assessment