All Articles
A Japanese query won't surface its English twin — when embeddings notice language before meaning
Embed a translation pair with gemini-embedding-2 and the two halves won't be nearest neighbours, because language itself inflates similarity. Here is how I measured cross-lingual recall using translation pairs as ground truth, and what happened when I subtracted the language centroid.
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.
A near-miss label won't fix itself on retry — a normalization layer for closed-vocabulary classification
When responseSchema enum returns an out-of-set label, retrying tends to return the same near-miss. From a wallpaper app's 30-category batch, here is the distribution of how labels miss, plus a normalization layer built on an alias table and gemini-embedding-2 nearest-neighbor, with measured results.
Spend Deep Reasoning Only Where It's Needed: Per-Request thinking_level Routing in Gemini
Running every request at high thinking_level bloats latency and cost; forcing low drops accuracy on hard questions. This walks through a router that picks Gemini 3.x thinking_level per request from an inexpensive difficulty estimate, keeping p95 latency inside a mobile budget while reserving deep reasoning for the questions that need it — with measured numbers and working code.
A minimal evolutionary search loop with Gemini: propose, evaluate, select — prompted by AlphaEvolve's GA
With AlphaEvolve reaching GA, I built the smallest possible evolutionary search loop on the Gemini API: generate candidates, score them with a fitness function, and select the best. Sandboxed evaluation, diversity, and budget control — from real solo-dev use.
A Risk-Tiered Approval Gate for Gemini Function Calling
Handing full autonomy to an agent is unnerving. This walks through a Gemini function-calling loop that routes tool calls into auto-run and hold-for-approval by risk tier, then feeds the result back to the model after a human signs off.
The Day We Went From 30 Categories to 34 — Reclassifying 1,180 Assets Instead of 8,142
Adding categories to a taxonomy does not require reclassifying everything. Here is how embeddings and confidence margins narrowed a backfill from 8,142 assets to 1,180, with the numbers.
Images Made With a Retiring Model Can Never Be Made Again — Tracking Regenerability in a Ledger
When Gemini's image generation models shut down on August 17, the assets you made with them can no longer be reproduced the same way. Here is the ledger design and code I use to decide, before the deadline, which assets are regenerable and which must be frozen.
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.
When Your Knowledge Base Shifts Mid-Run: Pinning File Search to an Execution Epoch for Consistent Agent Grounding
When a File Search store is updated while a Managed Agent is running, a single execution can mix old and new grounding. Borrowing MVCC ideas, pinning an execution epoch keeps one agent run's evidence consistent. Here is the design and implementation.
One File Search Store for Many Apps: Splitting Retrieval With customMetadata and Chunk Config
Put several apps' FAQs in a single Gemini File Search store and metadataFilter can silently return empty grounding, or answers get split across chunk boundaries. Here is the customMetadata design, the AIP-160 filter-syntax trap, and measured chunkingConfig tuning.