The Process of Exploration in AI Research: A Researcher’s Perspective
Talkat Open Source Summit India 2026 · on Wednesday, 2026-06-17· at 15:35· Jasmine Hall 2 (Level 3)·
summary
Describes Pile of Flow, an open‑source local experiment manager that converts code blocks into component-as-dictionary entries, centralizes configs, logs and artifacts, and runs everything from a fixed Jupyter notebook workflow. Claims to collapse the combinatorial explosion of model/data/optimizer choices, make per‑component auditing possible, and speed iterations while saving compute. Project is implemented (GitHub: Pile of Flow), evolved from a paper, and is presented as usable via ExperQuick.org.
AI-written summary · about
chapters (13)
- 0:00AI research exploration
- 1:04AI research misunderstood
- 2:05Iterative AI research process
- 4:55AI research experiment
- 8:05Experiment components complexity
- 12:36Time for experiment understanding
- 15:29Experiment tracking challenges
- 20:02AI researcher responsibilities
- 21:51Experiment auditing problem
- 23:17Querying experiment spaces
- 25:04Pile of Flow solution
- 28:15Future of AI research with Pile of Flow
- 29:11Q&A
related talks
01h 29mFull-stack Machine Learning and Generative AI for Data Scientists
29mAI/MLOps for Busy People: A Field Guide To Implementing Cloud Native AI/ML
30mMLflow: an Open Source Tool for Managing the ML Project Lifecycle
25mBuilding AI Workflows: From Local Experiments To Serving Users
25mDirecting a Swarm of Agents for Fun and Profit
40mEnd-to-end AI Marketplace with Execution Engine on Kubernetes
53mIntroducing AI to the Flowable Process Engines
41mFrom Lab To Life: Practical AI System Evaluation