Abstract

For about five years now, the real bottleneck has not been infrastructure; and yet, most organisations have run Spark or Glue jobs for datasets that comfortably fit in a single machine’s memory. We believe the real villian has been tooling and we are here to fix that.

Bringing our past experience together, we built OrcaSheets, a powerful desktop application that lets analysts build full ETL workflows locally, with no cluster provisioning, no cloud egress, and no YAML configs. With the infrastructure problem solved, the remaining challenge was adoption: our internal DSL for transformation pipelines could not be exposed directly to users, and the unlearning curve had to be smooth. So we built an agent that generates agents.

OrcaSheets unbundles the warehouse in a fixed order. Compute moves to a local platform for the jobs that fit in memory. Storage stays at the source behind connectors. ETL infrastructure becomes a versioned recipe. And finally a recipe generator: English in, a validated agentic workflow out. And finally we build the data lake for teams that don’t have one.

I will demo the path on a large local database: a real use case, then a deep dive into the recipes. Figures we share are from that class of workload: on the order of seconds for a million-row file, zero egress when data stays on the machine, and roughly 10–50× cheaper than equivalent Glue or Spark jobs. What failed is part of the talk: we built a DSL that analysts would not learn, and we would have shipped a weaker product if we had put the agent on the warehouse first.

Takeaways

1. A practical framework for when local-first analytics replaces cloud ETL (Spark/Glue), with real cost comparisons showing 10-50x savings.
2. How AI-generated recipes reduce pipeline development time from days to minutes while keeping data fully on-premises.
3. Unbundle compute from storage, build agentic workflows, and then finally an agent that generates agents for the user who will not learn the language.

Target Audience

• Data engineers exploring alternatives to cloud ETL for mid-size datasets;
• ML engineers interested in local-first analytical tooling;
• Platform engineers building internal DSLs or agent-powered developer tools;
• Anyone building AI agents that generate structured, validated outputs rather than freeform text.

Customers

OrcaSheets is being used by a major Government in the UAE, one of the largest conglomerates in India, a listed phramaceutical manufacturer in India, D2C brands and about 5000 Daily Users.

Bio

Yash is the co-founder at Orcasheets. He has in the past built practical, scale-proof systems, starting from early days of Machine Learning Productionization to building Highly scalable Engineering systems as an Engineering Leader to now leading Product Management at OrcaSheets. A three time founder, Yash also has two patents in Short Text Language Models and Intent Based Routing Systems.

At OrcaSheets, we are making analytics that run locally on your PC, handling 100M+ rows in seconds and turning raw data into insights, dashboards, and decisions.

Link

https://drive.google.com/file/d/1ETOf7NXKAeH0bF_YE_NIMcUvQT2Kv82h/