Alterra Answers features: - Ask questions using natural language - Search by meaning, not just by keywords - With up to 90% precision, it is the most accurate question answering bot in the industry - Deep Learning inside – no coding required More >>

Throughput and latency. Alterra Answer’s and FAQ API’s response time (latency) is < 300 ms. One medium AWS server can hold up to 1,000 queries/sec. We can always add more servers.

You can mute your bot. Just press the Mute button in the right upper corner of the FAQ Editor. In the mute mode the bot will process the user queries and log the results but not show any replies to users. You can see the logged results in the Label tab. You may wish to use the mute mode before you publicly launch your bot, for quality testing purposes, or whenever you want it to stay silent, e.g. when live agents are on duty. ...

What if the bot doesn’t know the answer? Will it stay silent? Well, it depends. In general, the bot tries to find the most likely answer in your FAQ, and reply. So, by default, it always replies. ...

All Alterra products are available worldwide. However, currently, the system supports only English. 100 more languages are in Beta. Contact us for the details.

The bot always gives one answer to a given question. Many questions can have a common answer, but identical questions always lead to identical answers. In fact, these answers are to be written by you. The bot finds the right answer in your FAQ. There shall always be only one right answer there.

BERT (Bidirectional Encoder Representations from Transformers) is a Deep Learning - based language model developed and open-sourced by Google Research. It is a bidirectional, unsupervised language representation, pre-trained using a plain text corpus (Wikipedia). It makes very good word embeddings for 102 languages that we use as part of our algorithms. More here

Alterra Answers can work with arbitrarily large knowledge bases, long articles (answers) and long queries (questions). Whereas FAQs / knowledge bases typically don’t have more than a few hundred articles, the bot can handle tens of thousands. The article length is not a problem either. Likewise, the bot can handle queries from just one or two words to relatively long ones, consisting of a number of sentences.

The bot includes a built-in spell checker, so it tolerates typos

The TRAIN button activates the process of training the bot. After you have uploaded your knowledge base, labeled new queries or edited a FAQ article, make sure you pressed the TRAIN button! Without it, your edits won't take effect. Every Machine Learning algorithm has a training stage when it learns from the examples you gave it. It may take the bot a minute or so to digest the new examples. ...