Interview with Ankur on the DSPy Interview Series podcast
| Time | Topic |
|---|---|
| 0:00 | Introduction |
| 0:43 | Origin story and background |
| 1:35 | Why Ruby needed DSPy |
| 3:56 | BAML adoption and benchmarks |
| 6:56 | Tune format discussion |
| 9:39 | Porting strategy decisions |
| 11:35 | Database migrations for prompts |
| 15:02 | Project work and consulting |
| 19:14 | Ruby LLM integration |
| 24:50 | Agentic workflows vs agents |
| 27:21 | DSPy.rb roadmap |
| 31:01 | AI industry predictions |
“I had two options: either implement Rails in Python - and Rails is like 20 years old with thousands of contributors, I think that’s impossible - or try to surgically port parts of DSPy the Python library into Ruby.”
“BAML really shines when you have a rich signature - like you have a deep input schema and an output schema with objects. That’s where the savings come.”
“What I’m doing is working on an interface around optimizers to manage optimized prompts as if they were database migrations… you keep track of migration files, you run the migration, and you’ve got a file with the schema as the source of truth.”
“There’s still a lot of people conflating agents and workflows, conflating the need for an agent when they actually just need a workflow, maybe a three or four step workflow.”
“I can think strategically. I can zoom out as a software engineer… I’m building a cathedral and I can actually get in my drone and look at people using the cathedral.”
Vicente Reig - @highwayvaquero