Lessons Learned Creating a Successful Machine Learning Startup with Clojure
summary
Founded in 2011 using Clojure for web and data preprocessing, the company scales from a 3‑person prototype to 40+ engineers and later over 150 staff while using Java/C libraries for machine learning and OCR. Development moves frontend from CoffeeScript to ClojureScript and Reagent, adopts Clojure for web and preprocessing, introduces Go for core data processing to reduce GC and JNI memory fights, shifts serialization to Protocol Buffers for large data and documents memory overhead from immutable data structures.
AI-written summary · about
chapters (24)
- 0:00Introduction to Kira Systems
- 2:07Company founding principles
- 3:33Why choose Clojure
- 5:18Initial tech stack choices
- 7:42Early stack evolution
- 8:04Great things about Clojure
- 10:51Challenges scaling with Clojure
- 14:02Readability vs conciseness
- 14:23Frontend library struggles
- 15:51Memory overhead surprises
- 17:37Adopting schema and spec
- 19:21Testing pitfalls in REPL
- 20:25Future tech stack changes
- 23:12Current production stack
- 24:39Lessons for Clojure community
- 27:46Improving tooling and testing
- 29:29Q&A introduction
- 29:51Legacy code and team growth
- 31:13Velocity with larger teams
- 33:39Hiring and training new devs
- 36:23Learning Clojure timeline
- 37:06Serialization framework choices
- 38:07Starting Kira today
- 39:52Closing remarks
related talks
17mBetting the company on Clojure
34mDeepen and Diversify the Clojure Community with Jr Engineers
35mTotal Massacre: Reducing the Middle Office
44mLearning and Teaching Clojure on the job at Amperity
44mData Science in Clojure
28mMachine Learning Live
29mDeep Learning Needs Clojure
37mTeaching Clojure at IBM