The companies and institutions we work with

Arcada is a United States software company — a New Mexico LLC — building products end to end: the engineering, the infrastructure, and the operational work that begins once something is live. It builds for its own portfolio and for companies in the network.

Several of the products listed here were built inside Arcada and then spun out to stand on their own: a customer-messaging platform, a live call translator, a legal document generator, a certification publisher and a trading terminal. That is the reason a group this small can be useful to a founder beyond the cheque — the people reviewing your architecture have shipped the same problem recently.

Small and mid-sized businesses talk to customers in a dozen places at once — Telegram, WhatsApp, Instagram, web chat, email, marketplaces. Requests get lost between them, nobody can see who answered what, and the conversation history walks out of the door with the employee who leaves. Western tools solve this, but they price for enterprises and several of them do not connect to the channels these businesses actually use.

Messello is a shared team inbox across all of those channels, with a light CRM underneath it — customer card, full history, notes, attributes — plus routing, templates, canned replies and AI-drafted answers. It can be self-hosted, which means the business owns its own conversation data rather than renting access to it.

It is aimed at DTC and e-commerce teams already living in WhatsApp and Instagram, and at small support teams who want the shape of Intercom without its price. Because it can be self-hosted and the channel connectors are open, it also reaches markets whose messaging channels the large incumbents never bothered to support.

Doing business across borders means taking calls in a language you only half control. Existing tools either transcribe after the fact, which is useless while you are speaking, or live inside one meeting platform, which is useless when the client picks a different one.

Subtl captures the other side’s audio from the system output and your own microphone on the device, then draws a translucent overlay above whatever application the call is happening in — Zoom, Meet, Teams, Telegram, WhatsApp or FaceTime. The overlay shows the original text, its translation, and a suggested answer drawn from documents you loaded before the call, so a pricing question gets answered with your pricing rather than a guess.

It is a desktop application rather than a browser extension precisely because the capture has to work regardless of where the call is. The engineering underneath is unusual for a product this small: a Rust client core for segmentation and streaming, a Swift library for macOS system-audio capture, and a Go realtime gateway handling speech recognition, translation and suggestion in one pass.

Every SaaS, mobile app and online store needs a privacy policy and terms of service, and most of them copy one from a competitor. That is a real exposure: the copied document names the wrong jurisdictions, misses the regulations that do apply, and describes data handling the business does not actually do.

Lextiva asks a structured set of questions about the business, works out which regulations are in scope from the answers, and generates a matching document set — privacy policy, terms, cookie policy, DPA, AI use policy — as HTML, PDF and DOCX. Templates carry the legal structure; a model fills in the specifics of that particular business, so the output reads like it was written for them rather than found.

The audience is deliberately the one that lawyers price out: indie founders launching a product, small agencies shipping client sites, and merchants on Shopify or Webflow. Not a substitute for counsel on a hard question — a way for a small company to stop being non-compliant by default.

Certification prep splits into two bad options. Cheap question dumps copy real exam items and teach nothing. Expensive courses go stale the moment the vendor updates the exam. In both cases the buyer cannot tell which questions are current or even correct, and wrong answers are the single largest driver of one-star reviews in the category.

EducatedBear writes each course from the vendor’s published exam outline and states which version of that outline it was built against. Drafts are produced with AI and then re-solved by an independent reviewer who does not see the answer key; every source link is opened and checked, and anything that fails either pass is held back rather than shipped. Every question explains why each option is right and why the others are wrong.

A course is one payment with permanent access and free updates when the outline changes — no subscription. The first release runs to 421 pages with 267 verified questions, a final mock exam, lab walkthroughs and a command reference, delivered as watermarked PDF and EPUB alongside an unlimited browser-based practice trainer.

EducatedBear is independent and is not affiliated with, authorised, sponsored or approved by any certification vendor.

Prediction markets have grown into serious venues, but the tooling around them has not kept up. Traders running real size on Polymarket work with basic charts, no alerting worth the name, and none of the order types a professional would expect.

Polyterminus sits on top of Polymarket and adds the missing layer: advanced charting, smart alerts, AI assistance on positions and markets, copy trading, and emulated order types the underlying venue does not offer natively. The first phase targets traders who are already active there, linked by importing an API key manually.

It never custodies funds. USDC stays in the trader’s own wallet exactly where it was, and the terminal only reads and routes — which is both the correct design and the only version of this product that does not become a regulated money business.

A usable portrait is a small, permanent problem. Most people solve it badly — an old photo, a bad crop, a studio session nobody books — and the generative tools that appeared to fix it largely live on the desktop web, charge before showing anything, and return a style the user did not choose.

SnapMe is a phone app. A user supplies a few reference photos, and the engine trains a lightweight model on their own features before generating: portraits come back in under forty seconds, with background replacement, relighting and depth separation handled automatically, and a one-tap export sized for LinkedIn and the other places a profile picture actually goes. Generation is metered, with a free daily allowance, so the first result costs nothing and the pricing decision comes after the user has seen the quality rather than before.

It runs on iOS and Android from a single React Native codebase, with a Python inference tier on GPU instances behind it — a fine-tuned Stable Diffusion XL pipeline optimised for faces rather than a general image model pointed at people. That distinction is most of why the output holds up at portrait crop.