What language should your system instruction be in? Measuring three approaches when most prompts arrive in the user's language
Keep the system instruction in English, or translate it into the user's language? I measured input tokens per language with countTokens, then lined up output-language match and schema compliance to find where nine tokens is enough.
I Asked Gemini to Grade My App Store Screenshots. Everything Scored 78–85.
Ask Gemini Vision to grade App Store screenshots out of 100 and good candidates and deliberately broken ones both land at 78–85. Here is how I measured the judge's discrimination power, dropped absolute scoring, and rebuilt it as debiased pairwise comparison.
Sample what you already accepted — an audit budget that catches silent quality drift
A confidence gate only ever looks at output the model hesitated on. Silent drift sinks into the batch that sailed through. Working from a fixed 30-minutes-a-day review budget, this walks through deriving detection time from the binomial, reallocating the same budget across risk strata, and catching slow decay with a cumulative monitor.
When Gemini's executed result and its prose disagree on a number — a gate that trusts only code_execution_result
Gemini Code Execution returns the value it actually computed and the sentence describing it as separate parts. Trust the prose and you can inherit a hallucinated number. Here is a verification gate, in working code, that extracts the executed result as the single source of truth and rejects prose that disagrees.
When responseSchema Can't Do $ref: Handling Recursive Schemas in Production with responseJsonSchema
Gemini's responseSchema is an OpenAPI subset with no $ref or $defs, so it can't express shared definitions or recursion. Here's how I moved to responseJsonSchema to reuse localized fields and handle a recursive category tree in production.
When a Whole Chapter Vanished From My Long-PDF Summary — Field Notes on Auditing Coverage
Chapter-wise summarize-then-merge works well on long PDFs, but a lost extraction or a merge step can drop an entire chapter while the pipeline finishes without a single error. This walks through catching those silent drops with one number—coverage—and re-running only the chapters that fell out.
My ADK Assistant Quietly Forgot a Deadline — Catching Compaction Memory Loss With a Recall Probe
Compacting conversation history in Google ADK with Gemini lowers cost, but it also erodes what your assistant remembers — silently. Here is how I built a recall probe to measure that loss, compared three compaction strategies against the same ledger, and stopped trading memory for tokens.
Setting a Token Budget Per Free User: Balancing AdMob Revenue Against AI Feature Cost
Rate limits protect requests per minute. They do nothing for the invoice that arrives at the end of the month. Here is how I derive a per-user token budget from ad revenue, keep the ledger inside a single call wrapper, degrade gracefully at a soft cap, and detect abuse with one concentration ratio.
Google Sheets API × Gemini API: A Python Data Pipeline — No Apps Script Required
Learn how to build a fully Python-based pipeline that reads data from Google Sheets, processes it with Gemini API, and writes results back — without touching Apps Script. Covers service account auth, structured output, and rate limit handling.
Being Able to Say Later Where a User's Data Was Processed — Region-Pinned Gemini (Vertex AI) Clients and a Residency Policy Gate
The default global endpoint is convenient, but it leaves you unable to explain where an EU user's data was processed. Here is a design built from three parts—a region-pinned Vertex AI client, a policy gate that forbids implicit global fallback, and a call ledger—with working code and measured latency.
When Gemini API Leaks Japanese Into Your English Output Once in a While — Field Notes on Measuring the Contamination Rate and Tightening It in Stages
You told Gemini to answer in English, and 3 out of 100 runs slip a Japanese sentence into the tail. Here is why you cannot stop that 'once in a while', and a production pattern that measures the contamination rate as an SLO and tightens it with graded recovery, with working code.
Catch Near-Duplicate Posts Before You Publish — a Topic-Cannibalization Gate with Gemini Embeddings
Once a blog passes a few hundred posts, new articles start cannibalizing old ones in search. This walks through a pre-publish gate that embeds each post's meaning with gemini-embedding-2 and blocks drafts that sit too close to something you already wrote — with runnable code and how to pick the thresholds.