Building a better eBay: HAWK
Last month Gravitywell set up shop in Mallorca for our annual hackathon.
We set out with a simple enough premise: build a better eBay.
Selling things, even on a mature platform like eBay is inherently time-consuming and resource intensive, but somehow listings end up being identikit copies with little or no customisation. In general, shopping online often feels like trying to compare things at five different car boot sales (next time you're shopping around for something try counting the number of tabs open in your browser).
Buying online has risks, even if you're dealing with an established company. How comfortable are you giving over your credit card details to anything that isn't Paypal or Stripe? And to a private individual? No thanks.
Consider a user journey where I want to buy a new laptop from eBay.
I navigate to eBay, load the site, search, filter results and compare the options available, bid or buy, get redirected to paypal, complete the payment and then back to a confirmation screen. Eeesh.
How could we make this whole process simpler? Consider the following service called Magic:
This US based service acts as a digital concierge and uses a blend of natural language processing and human intervention to fufill user requests.
What if selling or buying something was as easy as this? As easy as texting someone?
Our project goal then, is to build a natural language-based interface for peer-to-peer transactions.
We knew that this wasn't going to be a project we could solve completely in a single week, the platform we had in mind was a year's worth of work away at least. But we knew that we could lay some solid foundations during the hackathon and in effect produce a prototype that demonstrates the core offering: natural language commerce.
The maxim goes "Ship early and often", so that's what we did:
HAWK is the umbrella name for our group of hyper-intelligent bots that will help you buy and sell, peer-to-peer. The HAWKs will help you find what you're looking for, will send you alerts when they come up for sale, and can look up information for you.
HAWK is the answer to fragmented user interfaces, contradictory user flows and multiple device formats: it uses an interface that anyone with a smartphone is innately comfortable with: a chat window.
Which is all very well, but what does HAWK actually do?
In affect, it tries to understand what you've just typed out in the chat window. We use natural language processing to get a handle of the intent of what you've typed, but the context of the conversation allows us to be quite opinionated about things. For example, if you type "Bad, by Michael Jackason"' or "late 90s techno", HAWK can break apart the query and use those parts to make intelligent lookups for you.
For our first market (yep, we're just getting started) we decided to try HAWK at buying an selling vinyl, as the screenshot above suggests. We chose vinyl because it fits the bill as a market with granular product items and a wide range of taxonomies and prices.
But a simple lookup and selling service would be far too dry. So, we leveraged Wolfram Alpha, the self-styled "computational knowledge engine" of the web, to suddenly give HAWK access to almost every piece of quantifiably useful information on the web.
HAWK can use this superpower to tell you how heavy the earth is, or the distance between any two points on earth, but Wolfram Alpha also has a surprisingly detailed knowledge of artist's back catalogues and other useful trivia.
Check out Hawk in action here:
This is the kind of use case that we think HAWK can excel at; an encyclopedia of knowledge at your fingertips that lets you quickly hone in on the product you want.
My user journey earlier can be reduced to three definitive actions:
- Establishing intent (I want a laptop, where can I get one?)
- Matching intent to a solution (Find me the right laptop)
- Confirming and concluding (Process payment, close transaction)
These three actions gave us our game plan for the hackathon, and a roadmap for assembling our tech stack.
We'll post more tech-flavoured goodness on the approach we took to building the application soon, but checkout this hitlist of supporting tech:
- Meteor / node.js - core APIs
- Telegram - message transport layer
- Wit.ai - natural language analysis
- Stripe - payment handling
- Sendgrid - email
So when can you get your hands on HAWK? As we said before, ship and often, so we'll be letting HAWK loose for a public beta towards the end of April. As you've seen from the screenshots above we've come a long way in a relatively short space of time, but we're keen to put some more polish on the aspects of service peculiar to vinyl.
So stay tuned and bear with HAWK as she starts stretching her wings. Sorry, I made it through this far, couldn't resist.