Satwik Singh is the co-founder of Echelon and an applied AI engineer who helped build the knowledge and research systems behind Alice, 11x’s autonomous sales representative. His work gives digital workers access to company-specific information and makes their answers inspectable by the businesses they represent.
Singh began working across machine learning and infrastructure in 2018, contributing to speech-recognition software at the National Center for Supercomputing Applications and insurance-pricing models at State Farm. He subsequently worked on Twitter’s build infrastructure, machine-learning ranking systems at Facebook, and vehicle software and fleet infrastructure at Tesla.
At Meta from 2021 to 2024, he developed generative AI for advertising, including background-generation and image-expansion tools for Feed Ads, and contributed to Reality Labs infrastructure. He joined 11x as an AI lead in 2024, building Alice’s knowledge base, a research agent, sales-campaign sourcing systems, and credits infrastructure before co-founding Echelon in 2025.
Building a sales agent that understands its customer
Singh and Sherwood Callaway demonstrated Alice’s architecture at the 2025 AI Engineer World’s Fair. Its most distinctive elements include:
- Customer-specific knowledge base: Alice ingests sales decks, case studies, websites, recorded calls, and videos, replacing manual onboarding forms with materials companies already use to train salespeople. An asynchronous pipeline combines Amazon S3, document parsing, embeddings, and Pinecone retrieval.
- Markdown-aware chunking: Parsed documents are split first by headings, then sentences, then tokens, preserving meaningful sections while keeping retrieved passages compact enough to support grounded sales outreach.
- Deep-research agent: Instead of returning isolated search results, the agent plans multiple retrieval steps, gathers relevant evidence, and synthesizes answers. Singh brought the system into production using Letta; his account of the implementation emphasizes usable answers without requiring customers to understand vector databases.
- Customer-visible agent reasoning: Users can inspect supporting passages, interrogate Alice’s knowledge, and explore embedded documents through a three-dimensional visualization—practical safeguards when an autonomous agent represents a company to potential customers.
