HomeCompaniesCaptain

Self-tuning file search for AI agents

Captain syncs complex, multimodal files and cloud storage from sources like S3, SharePoint, and Google Drive, and makes that knowledge searchable for your agents. Instead of stitching together parsers, embeddings, vector databases, rerankers, and retrieval infrastructure, Captain self-tunes the retrieval pipeline to your data and workloads. On benchmarks, this improves accuracy from roughly 78% with standard RAG to over 97% by tuning retrieval to the unique structure and content of your data. Captain is purpose-built for complex, regulated, and high-stakes file search workloads. Learn more at https://captain.dev
Active Founders
Lewis Lien Polansky
Lewis Lien Polansky
Founder/CEO
CEO @ Captain, building self-tuning file search for AI agents. Engineer, designer, founder of a cybersecurity CTF org, recipient of U.S. Congressional recognition for community software, and randomly co-invented the soy-based milk carton.
Edgar Babajanyan
Edgar Babajanyan
Founder/CTO
CTO @ Captain | Published AI NLP Researcher | Built OCR engines @ Boar's Head | Scaled high-performance RAG pipelines for the past 4 years | Built a Cross Region L2 Datacenter
Company Launches
Captain - The Reliable Alternative to RAG
See original launch post

TL; DR: Captain delivers the most accurate general-purpose knowledge search engine ever built. 

If you need to search through large text or multimodal files, and:

  • Don’t have time for a tedious RAG sprint -or-
  • Your current search pipeline lacks accuracy

We should talk. → runcaptain.com/sales or founders@runcaptain.com

https://www.youtube.com/shorts/RxMPHqou94o

In a world where 90% of enterprise knowledge cannot be stored in traditional databases, this ‘unstructured data’ is an untapped goldmine for decision-making. 

The issue is that current RAG solutions have poor retrieval quality overall and are only performant on pre-optimized question types. 

uploaded image

Captain unlocks better responses thanks to an effectively Infinite Context window. We distribute it across many LLMs in parallel + some embeddings sprinkled in, and then ‘Map-Reducing’ responses down to a single output.

uploaded image

Without context limits, we have great freedom to optimize our retrieval engine for maximal accuracy. We can now play with a truly dynamic top-k or just run the LLM exhaustively (if a full knowledge audit is needed).

uploaded image

We’re two lifelong builders obsessed with data! 

  • Lewis (right) solved hallucinations for code generation through his previous startup. 
  • Edgar (left) researched NLP+AI and has built production RAG pipelines for the past 3 years.

We’ve seen the pains and limits of these systems from the start, and the industry is primed for a more accurate alternative. 

This past summer, we met with every engineer we could at Snowflake and Databricks, and we kept hearing the same thing again and again: 

There’s no good scalable unstructured data search.

Until today.

Captain abstracts the engineering of retrieval entirely. Connecting files is all that’s needed, and we’ll beat your RAG's accuracy. Guaranteed. 

Our Ask

If you know a CTO or Head of AI at a mid-market or enterprise AI-native (or are building one yourself), we would love an introduction!

Check us out at RunCaptain.com
or reach us directly through founders@runcaptain.com

Happy Shipping! 🚢

YC Photos
Captain
Founded:2025
Batch:Winter 2026
Team Size:2
Status:
Active
Location:San Francisco
Primary Partner:Garry Tan