Slide 1

Slide 1 text

Arnaud Bailly [email protected] http://www.capital-match.com @abailly 1 Haskell All the Way Down Building a startup using Haskell

Slide 2

Slide 2 text

Functional Programmers Paris - 2015-04-01 Agenda ▪ Who/Why/What ▪ How we use Haskell ▪ The Good, the Bad and the Ugly ▪ Future work ! This is not a Monad tutorial…. 2

Slide 3

Slide 3 text

Functional Programmers Paris - 2015-04-01 Who? What? Why? 3

Slide 4

Slide 4 text

Functional Programmers Paris - 2015-04-01 Who? ▪ Pawel Kuznicki – CEO, ex-Rocket, ex-McKinsey, ex-Zalora ▪ Kevin Lim – CFO, ex-JPMorgan, ex-SCB ▪ Arnaud Bailly – 20+ years experience, mostly Java for food, discovered Haskell in 2001 ▪ Willem van den Ende ▪ 20+ years experience, XP since the beginning, Smalltalk/Clojure/.Net… 4

Slide 5

Slide 5 text

Functional Programmers Paris - 2015-04-01 ! What? Marketplace Lending! ~S$32,400 repayment in 1 year Bank P2P lending Bank n lenders n deposits = S$30,000
 @ 1.5% interest rate SME(borrower) S$30,000 loan @ 8% ~S$30,450 repayment in 1 year Return of 6.5% or S$1,950 ~S30,500 or 1.5% RETURN 5 Capital Match Lender 1 Lender 2 Lender 3 SME(borrower) Loan syndication on behalf of the lenders S$30,000 loan @ 6.3% 1%+ underwriting fee = S $300-400 S$5,000 loan @ 7% S$5,000 loan @ 7% S$20,000 loan @ 6% ~S$31,900 repayment in 1 year ~S$31,500 repayment 20% commission on returns = S$400 ~S20,960 or 4.8% ~S$5,280 or 5.6% ~S$5,280 or 5.6% 1% provisional fund = S$300 Provis. Fund

Slide 6

Slide 6 text

Functional Programmers Paris - 2015-04-01 Why Haskell? ▪ Pawel had very good experience working with Haskell developers at previous job ▪ He posted job offer on http://functionaljobs.com ▪ I wanted to do some real stuff in Haskell ▪ I had good experience working for people in Singapore ▪ It seems fun! Let’s do it! 6

Slide 7

Slide 7 text

Functional Programmers Paris - 2015-04-01 How we use Haskell? 7

Slide 8

Slide 8 text

Functional Programmers Paris - 2015-04-01 (Nearly) Everything! ▪ Dev. Env ⟶ ghc-mod, stylish-haskell ▪ Web Backend ⟶ Scotty, Blaze ▪ Database ⟶ Custom Event Sourcing ▪ Unit/Integration Testing ⟶ HSpec, QuickCheck ▪ End-to-End Testing ⟶ hs-webdriver ▪ Build ⟶ Cabal, Shake ▪ CI Server ⟶ Bake ▪ Configuration Management ⟶ Propellor 8

Slide 9

Slide 9 text

Functional Programmers Paris - 2015-04-01 For the rest… ▪ Web Front-end ⟶ Om/Clojurescript ▪ Version Control ⟶ Git (what else?) ▪ Packaging & Deployment ⟶ Docker (because we can) ▪ Infrastructure ⟶ DigitalOcean / S3 ▪ Monitoring ⟶ Riemann, collectd (WIP) 9

Slide 10

Slide 10 text

Functional Programmers Paris - 2015-04-01 Emacs Dev. Envt. 10 Compilation FlyCheck HLint Autocompletion

Slide 11

Slide 11 text

Functional Programmers Paris - 2015-04-01 Scotty: REST Endpoints 11

Slide 12

Slide 12 text

Functional Programmers Paris - 2015-04-01 Scotty: Middleware 12

Slide 13

Slide 13 text

Functional Programmers Paris - 2015-04-01 Blaze: Template HTML 13

Slide 14

Slide 14 text

Functional Programmers Paris - 2015-04-01 HSpec: Integration Tests 14

Slide 15

Slide 15 text

Functional Programmers Paris - 2015-04-01 QuickCheck 15

Slide 16

Slide 16 text

Functional Programmers Paris - 2015-04-01 hs-webdriver: ETE Testing 16

Slide 17

Slide 17 text

Functional Programmers Paris - 2015-04-01 Shake: Better make 17

Slide 18

Slide 18 text

Functional Programmers Paris - 2015-04-01 Bake: CI 18

Slide 19

Slide 19 text

Functional Programmers Paris - 2015-04-01 Propellor: Config Mgt. 19

Slide 20

Slide 20 text

Functional Programmers Paris - 2015-04-01 The Good, the Bad and the Ugly 20

Slide 21

Slide 21 text

Functional Programmers Paris - 2015-04-01 The Good ▪ Safer programming (shines in comparison with front-end dev) thanks to typing and compilation ▪ Types really help a lot: Documentation, intention, design, checking… ▪ Libraries and tools are most often good or very good even when in “beta" or “alpha" (e.g. bake) ▪ Nice and supportive maintainers and community ▪ We feel productive and confident to ship haskell code: Static Typing + Tests Rock! 21

Slide 22

Slide 22 text

Functional Programmers Paris - 2015-04-01 The Good (contd.) ▪ Refactoring is easier: Change a type and fix compiler’s errors ▪ Good for hiring: Haskell attracts “interesting" people ▪ Very easy to replace clunky scripts with typesafe and compiled DSL 22

Slide 23

Slide 23 text

Functional Programmers Paris - 2015-04-01 The Bad ▪ Cabal – but does its job, no binary packages possible and there is Shake for funky stuff ▪ Dev. Env. is still not on par with Eclipse/IntelliJ/ VS – but FPComplete and others are making progress fast and tooling improves ▪ Compilation typing errors – but you get accustomed to it once your code base is stable 23

Slide 24

Slide 24 text

Functional Programmers Paris - 2015-04-01 The Bad (contd.) ▪ Hiring: Hard to do if you require local people, Haskell communities are usually small. Be prepared to work remotely ▪ Can get pretty abstract pretty quickly… – pair programming and peer reviews to the rescue! 24

Slide 25

Slide 25 text

Functional Programmers Paris - 2015-04-01 The Ugly ▪ String vs. Data.Text vs. Data.Text.Lazy vs. – ⟶ Oh My! Haskell is Old! ▪ Runtime error reporting – ⟶ No Stack Traces! ▪ Aeson deserialization errors – ⟶ Cryptic No Parse ▪ Conflicting GHC versions/Libs requirements ▪ Lazy I/O can bite you 25

Slide 26

Slide 26 text

Functional Programmers Paris - 2015-04-01 Future Work 26

Slide 27

Slide 27 text

Functional Programmers Paris - 2015-04-01 Platform ▪ Replace clojurescript with Haskell based React bindings ⟶ ghcjs ▪ Generate CSS ⟶ Clay ▪ Replace Scotty with Servant for typesafe routes ▪ Improve End-to-End testing ▪ Performance and resilience testing ⟶ jepsen ▪ Distributed micro-services ⟶ Raft, Cloud Haskell 27

Slide 28

Slide 28 text

Functional Programmers Paris - 2015-04-01 Tooling ▪ Improve CI: – handle multiple branches, parallel builds, better reporting ▪ Improve configuration management – replace scripts w/ Haskell code for provisioning, better propellor configs ▪ Better containers builds and orchestration 28

Slide 29

Slide 29 text

Functional Programmers Paris - 2015-04-01 Credit Risk Analysis ▪ Build a database and inference engine to better automate Credit Risk Assessment process – NLP, ML, Big Data… 29

Slide 30

Slide 30 text

Functional Programmers Paris - 2015-04-01 30 We are hiring! \ http://functionaljobs.com/jobs/8802-full-stack-software-engineer-at-capital-match

Slide 31

Slide 31 text

Functional Programmers Paris - 2015-04-01 Questions? 31