Sony Arouje

At 4Mans Land we keep struggling to track our expenses. We normally note the expenses in Whatsapp and later move to a shared excel sheet. But this is always have difficulties, there will be lot of noises in the whatsapp with ongoing discussions. We also use slack for discussions, so I decided to integrate an expense app and evaluated some. Most of them are paid and don’t want to pay for an app at this stage. One night decided to write an app for slack and by morning finished the basic app that can store expenses in a mongodb. This whole experiment started as a fun to play with some thing new and also understand the slack api.

Wrote the server in nodejs and choose mongo db for persistence. For testing the slack integration, I used ngrok to create a local tunnel. also evaluated localtunnel.me which is a free version, but very unstable. ngrok is good but we have to pay $5 per month for license.  Started evaluating other alternatives to host my server, heroku was another alternative and used it several times earlier. At the moment we wanted to spend less for hosting and infrastructure. At last decided to host the server at my home, I have a really good stable internet connection with more than 80mbps speed and a Gigabit router. Added a port forwarding at my router,  expected to access the port immediately but to my surprise it’s not working. Done port forwarding so many times earlier and could access stuffs inside my network without any issues, my ISP was different that time. Then called up my ISP support, they informed me that without a static ip I wont be able to do port forwarding, for static IP I have to pay a very minimal fee and is life long unless I switch the provider. Paid the amount and in 2 days got my static IP and could do port forwarding successfully. Also set up dynamic dns at noip.com to give a good name to my IP. With all these settings done, changed the url in slack app and used the one I setup at noip.com. Run the nodejs server at my dev machine and fired a command from slack app, viola received the command to the server running in my laptop. The server is ready and is running in my laptop, but required a system which could run 24/7. The cheap option came to my mind was to host the nodejs server in one of the Raspberrypi lying in my table, one night I decided to setup the pi with the latest Raspbian stretch.

Setting up the Raspberry pi

Expected to finish the whole setup in less than an hour, as did this several times earlier but not that day. After installing the stretch, apt-get update started throwing hash sum error, reinstalled the os couple of times. Same error again and again. This error sent me for a wild goose chase, reading one forum after another and tried all the suggestions but nothing changed. At last came across this stackoverflow post which fixed the issue and could update the os, phew.

Next installing the nodejs, used nodejs version maintained here and issue the ubuntu command, for e.g. to install nodejs 10 issue the below command from your raspberry pi

Next task is installing Mongodb, apt-get install mongodb installed an earlier version of mongodb and is not suitable for the mongoose orm which I am using in the code, either I have to use an older version of mongoose or try to find a way to install a new version of Mongodb. I choose the later and try hunting for some clues to install some newer version of mongodb. At the end, come accross this post, which has all the details of installing mongodb in Rpi jessie, which is good enough in stretch as well, also he provided a link for stretch mongo binaries. Followed the instructions and at the end mongodb started running in my machine.

When I try to connect from Compass to rpi’s mongodb instance, it’s not getting connected. Reading through some docs and forums realized that I have to comment bind_ip in /etc/mongodb.conf, commented out that config and able to connect from compass.

Before going to bed at 2am, copied the nodejs code to rpi, did some testing and all went well. Could post expenses from slack to the service running in my pi. Went to bed with peace of mind.

What this app does?

The main focus of this app is to capture the expense, for now it uses slack command to forward text to the service. Initially created a strict pattern to record expense for e.g.

/expense spent 4000 INR on Jill to fetch a pail of water by Jack

Here spent should be first word followed by amount, after ‘on’ specify the description and after ‘by’ specified who paid the amount. The nodejs service will parse them and place them in separate fields in mongodb collection. Then to query the expense, some thing like

/expense query date today

/expense query date 05/12/2018 – 05/14/2018

So here ‘spent’ and ‘query’ is a keyword, based on this keyword different handling logics will kicks in and complete the task.

I felt the syntax is more restrictive, started analyzing options to process the sentence more naturally. I come across natural, a NLP library for nodejs. Using natural’s BayesClassifier trained the system to categorize the input text and derive the right keyword, which again calls the different logic and get the task done. After the classification training, the system can take inputs like

/expense Jack paid 4000 by Card for ‘Jill to fetch a pail of water’ on 05/16/2018

The above code will classified as spent, then added some code to extract relevant part from the text. Not pure NLP approach, any thing inside single quotes will be considered as expense. Any date in the text is considered as the payment date and so on. I am learning NLP, in future I might achieve a better translation of text.

Querying command is modified as shown below

/expense today

/expense between 05/12/2018 – 05/14/2018

Can also query and return expenses in csv format

/expense today csv  will return the expense in comma seperated format.

Slack output

After each expense created successfully, it returns the expense with the ID and the actual text we passed to the expense command. For e.g to create first expense passed the text like

/expense Jack paid 4000 to ‘Jill to fetch a pail of water’

Here is the output for querying todays expense, issue a command like

/expense today

In csv format (/expense today csv)

We can also delete an expense, only the expense you created can be deleted.

/expense delete today: delete all the expense you entered today

/expense delete 116: delete the expense with id 116

Had a great fun during this experiment and learned few things.

Happy coding…