TLDR: eSapiens is an AI-as-a-Service platform that securely connects Large Language Models (LLMs) with proprietary business data and workflows. It offers in-house control, no-code AI agent creation (Sapiens), and advanced retrieval-augmented generation (RAG) capabilities for both unstructured documents (DEREK engine) and structured databases (THOR agent). The platform emphasizes security, auditability, and ease of use, demonstrated by improved factual alignment in AI outputs and significant operational efficiencies for early adopters in high-stakes domains like legal and finance.
In the rapidly evolving landscape of artificial intelligence, businesses are constantly seeking ways to harness the power of large language models (LLMs) while maintaining control over their proprietary data and workflows. A new platform, eSapiens, emerges as a comprehensive AI-as-a-Service (AIaaS) solution designed to bridge this gap, offering secure and auditable retrieval-augmented generation (RAG) capabilities.
eSapiens is engineered to give businesses full command over their AI assets, ensuring that AI knowledge retention and data security remain in-house. It integrates seamlessly with proprietary data, existing operational workflows, and major agnostic LLMs, including OpenAI, Claude, Gemini, and DeepSeek. The platform empowers teams through its AI Agents, known as Sapiens, which are designed to provide valuable insights and automate repetitive tasks, allowing employees to focus on higher-impact work.
A core strength of eSapiens lies in its robust technical foundation. It features structured document ingestion, a hybrid vector retrieval system, and no-code orchestration powered by LangChain. A standout component is the THOR Agent, specifically built to handle structured SQL-style queries, transforming natural language questions into actionable insights from enterprise databases.
The effectiveness of eSapiens has been rigorously evaluated through two key experiments. A retrieval benchmark conducted on legal corpora demonstrated that a chunk size of 512 tokens achieved the highest retrieval precision, with a Top-3 accuracy of 91.3%. Furthermore, a generation quality test using TRACe metrics across five different LLMs revealed that eSapiens delivers outputs that are more consistent with the provided context, showing up to a 23% improvement in factual alignment. These results underscore eSapiens’ capability to enable trustworthy and auditable AI workflows, particularly in high-stakes sectors like legal and finance.
Described as an enterprise-grade AI “last-mile delivery” platform, eSapiens simplifies the integration of cutting-edge LLMs into real-world business operations. It features a unified data fabric capable of ingesting various document types, including PDFs, Office documents, and SQL tables. The DEREK engine is central to grounding every answer in verifiable context, while a visual no-code workflow builder allows non-technical staff to orchestrate complex multi-step actions using natural language.
Security is paramount for eSapiens, with enterprise-first measures such as end-to-end AES-256 encryption, fine-grained Role-Based Access Control (RBAC), and flexible deployment options including Virtual Private Cloud (VPC) or on-premise setups. SOC 2 Type II–aligned audit logging ensures compliance with data-residency and regulatory mandates, making it suitable for demanding workloads in finance, insurance, and life sciences.
The platform also boasts a marketplace with over 100 reusable prompts and 65 Quick Actions, covering diverse areas like customer service, sales, finance, marketing, and investment analysis. Early adopters have reported significant operational improvements, such as a reduction in monthly financial reporting time from two hours to twelve minutes, a 40% increase in automatic ticket categorization accuracy, and double-digit improvements in lead-to-deal velocity. These outcomes highlight eSapiens’ rapid time-to-value and sustainable operational impact.
eSapiens directly addresses critical challenges in the enterprise AI space, including data fragmentation, high integration and operational costs, limited trust and explainability, and talent gaps. By providing a unified data fabric, a citation-aware DEREK engine, and a workflow designer within a zero-trust security framework, it offers a consolidated value proposition for large enterprises and mid-market customers.
The platform’s core functional modules include:
Sapiens Creation
This module enables the creation of production-ready AI agents (Sapiens) through role-based modular design, no-code customization, and deployment, along with domain-specific persona templates. This allows for rapid agent assembly and deployment with built-in performance monitoring.
Prompt Management System
eSapiens offers a multi-level prompt library for reusable instructions, categorized by visibility scope (public, team, personal). It also includes Quick Actions, which are lightweight, single-purpose prompt snippets for high-frequency tasks, streamlining interactions and ensuring auditability.
Data Connectors & Knowledge Management
A unified data connector layer provides over twenty out-of-the-box connectors for various sources like Outlook, SharePoint, and Amazon S3. It ensures near-real-time synchronization, immutable versioning of documents, and intelligent retrieval through a hybrid vector index, supporting broad file formats and full auditability.
Security & Access Control
eSapiens implements a comprehensive, multi-layered security architecture. This includes secure transmission and AES-256 encryption, fine-grained access control, strong tenant isolation, continuous monitoring, and regular security reviews for LLMs and prompts. System-wide defenses like CAPTCHA, intrusion detection, and redundancy mechanisms further enhance security.
API & Embed Integration
Developers can extend platform capabilities via a secure RESTful API and embeddable UI components (JavaScript widget, iframe, React hooks). Pre-built adapters for third-party tools like Slack and Microsoft Teams allow seamless invocation of Sapiens and workflow outputs within native environments.
DEREK: Deep Extraction and Reasoning Engine for Knowledge
This module focuses on highly efficient and precise document question-answering. It preprocesses documents into chunks, uses OpenAI Embeddings for vectorization, and stores them in Elasticsearch Cloud. An intelligent recall workflow combines keyword and semantic searches, and answers are generated using optimized prompt formats (CO-STAR) and multi-agent validation to ensure high quality.
Also Read:
- CUE-RAG: Boosting LLM Accuracy and Efficiency with Advanced Graph-Based Retrieval
- Athena: Boosting LLM Accuracy Through Seamless External Tool Integration
THOR: Transformer Heuristics for On-Demand Retrieval
The THOR module is designed for structured data question answering. It automatically transforms natural language questions into executable SQL statements, interprets results, and provides actionable insights. Its decoupled architecture ensures scalability, with a self-correction module for query failures, making complex data accessible to non-technical users.
The system’s architecture is layered and modular, featuring a Knowledge Adaptation Layer that bridges diverse data sources with language model workflows, and an Application Logic Layer that orchestrates multi-agent workflows via LangGraph and LangChain. It supports major foundation models and integrates with third-party services, ensuring traceability and extensibility.
eSapiens is built on a multi-agent coordination framework, including a Data Analyze Agent for SQL-based queries and a Knowledge Base Agent for unstructured document stores. These agents leverage an enhanced foundation model, eSapiens-claude-3.7-extended, fine-tuned for long-context reasoning and high-precision language generation.
Practical applications include knowledge-centric question answering over enterprise archives, where users can retrieve policy or technical details from unstructured documents using natural language. Another key use case is natural language-driven structured data analysis, enabling business users to access key metrics from internal databases without SQL expertise. For more details, you can refer to the original research paper.
