When Vectors Break Down: Graph-Based RAG for Dense Enterprise Knowledge
AI Engineer World's Fair 2025 · 15:47
Enterprise AI agents and workflow automation
Writer builds an enterprise AI platform for creating, running, and supervising agents grounded in company data. WRITER Agent helps marketing, sales, and other business teams plan and execute work across company systems, while Playbooks package repeatable workflows. Developers can build applications through APIs and SDKs; nontechnical teams have no-code tools. The platform combines its Palmyra language models with company-data retrieval and administrative controls, and supports third-party models within WRITER Agent.
CEO May Habib and CTO Waseem AlShikh founded Writer in 2020, following their earlier work together on enterprise translation and localization. Its Knowledge Graph uses a specialized language model to identify semantic relationships, stores them in a graph, and compresses retrieved data with contextual metadata. Writer’s agent orchestration research also explores token efficiency through cache-stable prompts, incremental context compaction, and durable execution that lets interrupted work resume.
As of 2026, the company reported hundreds of enterprise customers, including KPMG, Intuit, Uber, and Vanguard. Sacra estimates Writer reached $47 million in annual recurring revenue in November 2024, up from $16 million in 2023. Writer closed a $200 million Series C in 2024 at a $1.9 billion valuation; its reported total venture funding is $326 million.
AI Engineer World's Fair 2025 · 15:47
AI Engineer Summit 2025 · 12:01
Affiliations reflect their AIE appearances, not necessarily current employment.
Start here to understand why strong general-purpose benchmark results may not predict reliable behavior in financial scenarios.
Waseem AlshikhAI Engineer Summit 2025
Learn how repeated terminology in enterprise documents can cause naive chunking and nearest-neighbor retrieval to return incorrect facts.
Sam JulienAI Engineer World's Fair 2025
Alshikh presents FailSafeQA to test financial models against misspelled queries, missing context, and OCR corruption. His comparison of reasoning and finance-specialized models emphasizes balancing answer robustness with grounding and declining to answer when evidence is inadequate.
Julien discusses graph-database scaling and Cypher limitations, then describes storing graph-derived data as JSON in a Lucene-based search engine and combining knowledge graphs with Fusion-in-Decoder.
Affiliations reflect each recorded session, not necessarily current employment.