Skip to content
galenry.
AI Labs

What we’re building before anyone asks for it

Internal R&D, kept honest about its stage. Some of this runs in production today; the rest is still finding out whether it should exist. None of it is public: this is what we are working on, not a product list.

The shelf

7 experiments · internal, not publicly available

Active
Beta
3
Working, in active hardening
Prototype
4
Exploratory, not supported, may never ship
How they work

The techniques underneath

Four patterns every experiment on this page is built on, whatever it does on the surface.

agent loop
model
tools
answer
max steps
8
max spend
$0.40
on limit
stop
The model calls tools, reads the result, and decides whether it has enough. The loop is bounded: it cannot run forever or spend without a ceiling.
evaluation harness
change
eval set
ship
revert
  • accuracy142 cases
  • refusals38 cases
  • latency p95under 2s
No prompt change ships without scoring against a fixed set. Regressions are caught before a user finds them, not after.
streaming response
0ms
180ms
410ms
620ms

first token · 180ms

Tokens reach the interface as they are generated. Perceived latency is what a user feels, not total time to completion.
guardrails
request
policy
execute
  • max spend / run$0.40
  • tools allowedread · search
  • irreversible actionneeds approval
Spend ceilings, allowed tools, and approval before anything irreversible. The model operates inside a boundary it cannot widen.

Beta

Working, in active hardening3

AI Website Generator

A brief in, a deployed marketing site out, with the design system already applied.

Next.js · codegen

AI Content Engine

Long-form drafting that holds a house style across a whole content set.

Claude · style embeddings

AI Document Intelligence

Extraction and classification at volume, with confidence scores you can route on.

OCR · structured output

Prototype

Exploratory, not supported, may never ship4

AI Image Studio

A spreadsheet of products or destinations in, production-ready cards, banners and social creatives out. Prompt templates per brand, generated in bulk.

Diffusion models · template engine · credits

Enterprise Knowledge Hub

Policies, SOPs and manuals indexed so staff get an answer with the source and page it came from, rather than a plausible paragraph.

pgvector · citations · per-tenant access

AI Proposal Analyzer

Reads several insurance proposals and compares premium, exclusions, waiting periods and claim conditions side by side, with the reasoning shown rather than just a verdict.

Document extraction · structured comparison

AI Voice Agent

Realtime speech with sub-second turn-taking, which is where most voice agents fall down.

Realtime API · WebRTC

Why none of this is linked

These are internal builds, most of them started from a client problem. We keep them unpublished until they are either a product we can support or a case study we can point at.

If one of them is close to what you need, we can usually get there faster from an existing experiment than from scratch — ask us about it directly.

Start

Tell us what you’re trying to build

An idea, a stalled project, or a system that has outgrown its architecture. Send it over and we’ll tell you where we would start and whether we’re the right team for it.

Response time
Within one business day
First call
30 minutes, no deck
Send us a briefhello@galenry.com

Prefer to write first? Send the brief and we’ll come back with questions before we come back with a proposal.