All Articles
Generate With Flash, Escalate to Deep Think Only When Unsure: A Two-Stage Pipeline
With Deep Think opening up on the API, the move is not to route every request through the heavy model but to have Deep Think verify only when Flash's output looks shaky. Here is the cost reasoning and working code.
Keeping Gemini API's Default-Model Shift From Becoming an Incident — Pinning Model IDs and Detecting Silent Upgrades in Production
When the default model quietly moves up, your output length, reasoning behavior, and cost change with zero code edits. This guide shows how to pin model IDs in a single source of truth and verify the effective model from the response to detect default changes.
Controlling Image Tokens with the Gemini API media_resolution Setting — Tuning Batch Image Classification by Measurement
media_resolution, introduced in the Gemini 3 line, switches how many tokens an image input consumes across three levels. Through real batch-classification measurements, this guide shows how to balance cost and accuracy by assigning the right tier per task.
When Gemini API Cuts Your Response Off Mid-Sentence — Detecting finish_reason: MAX_TOKENS and Stitching the Continuation
Long-form generation that ends mid-sentence is usually finish_reason: MAX_TOKENS. This failure arrives as a quiet HTTP 200, no exception. Here is how to detect it, stitch a continuation to recover the full text, and avoid the thinking-token trap that makes it worse on 3.x models.
Rebuilding a Three-Layer RAG Cache After Migrating to Gemini 3.5 Flash
When Gemini 2.0 Flash was retired, I rebuilt my RAG caching stack around 3.5 Flash. Here are the working implementations for response, semantic, and embedding caches, measured hit rates from production, and how self-managed caching divides the work with the API's Context Caching.
Where to Adopt Gemini 3.5 Flash GA First — Per-Workload Evaluation and a Staged Rollout with a Model Router
How I migrated production workloads to Gemini 3.5 Flash GA in stages: a per-workload evaluation harness, measured results, an env-based model router, and rollback design.
The Morning Gemini Generated Fine but the Publish Crashed — A 'Generation Outbox' So Expensive Output Is Never Lost
Generation succeeds, then the process dies right before publishing. The expensive output is gone, and you pay for the same generation again. Here is a 'generation outbox' that persists the output first and turns publishing into an idempotent follow-up, plus what it did for me during the June outage.
Reading a Night of Logs in Three Minutes — Building My Own Daily Brief for Ops With the Gemini API
Inspired by Gemini's Daily Brief, I built a pipeline that turns overnight operations logs into one morning email: collect, summarize with response_schema, render, deliver — with measured token counts and a fallback that kept working through the June outage.
Letting File Search's Multimodal Mode Find Wallpapers I Couldn't: A Field Report
I tested whether File Search's new multimodal retrieval (gemini-embedding-2) could replace category tags for finding one wallpaper among thousands. A 300-image trial, the walls I hit, and where semantic search actually fits — with working code.
Building an App Store Rejection Workflow with the Gemini API — From Structured Notices to Resolution Center Replies
How I use the Gemini API to parse App Store rejection notices into structured JSON, cross-check guidelines, draft Resolution Center replies, and run pre-submission checks as an indie developer.
Is Anyone Actually Using Your Gemini Feature? Measuring Acceptance, Regeneration, and Edit Distance
Token charts will not tell you whether users embrace a Gemini-powered feature. A practical design for measuring acceptance rate, regeneration rate, and edit distance with Swift and BigQuery, with two weeks of real numbers.
Retiring the Midnight Polling Loop — Rebuilding My Gemini Batch Monitoring Around Webhooks
A working log of migrating Gemini Batch API completion monitoring from 60-second polling to event-driven webhooks: static vs dynamic, signature verification, and real numbers.