install go: brew install golang

run: go run src/*.go

build: go build -o bin/*.go src/main.go
run binary ./bin/main

I've pulled in some data that q-bert generates, and theres a script to convert it to json documents that bleve can index.

node create-documents

When you start the go app, the generated documents will be indexed and an api stood up on 8094 with the following routes

• GET /api/fields get all the fields in the documents
• POST /api/search perform a search
• GET /api/qna/<document_id> get a document by its id

you can send requests from the postman collection

There are 2 indexes, which are combined into a single queryable index

One index has a simple text analyser that analyses all parts of the question so "what is your name" is indexed as ["what", "is", "your", "name"]

Another index has the standard bleave analyser, but with english stop word removal so "what is your name" is indexed as ["name"]

This approach helps guard against false positives for queries against a set of data with lots of stop word similarity, eg "what is your name" when we have indexed several "what is your [X]" items

Ref

https://gobyexample.com https://blevesearch.com https://github.com/blevesearch/beer-search https://godoc.org/github.com/blevesearch/bleve https://www.youtube.com/watch?v=CEfaIlzki5U http://bleveanalysis.couchbase.com/analysis

TODO

• dockerise
• frontend - web to speech?
• test suite checks accuracy.... bit like qbert...
• n-grams increase accuracy? of non-exact queries?