Enhancing Financial Analyst Efficiency Through RAG-Powered Search with Elasticsearch
Elastic helps financial analysts quickly extract relevant insights from dense and lengthy corporate filings for decisions making while including context and source references.
Empower financial analysts to efficiently extract key insights from complex and lengthy corporate filings - such as 10-Ks, 10-Qs, and annual reports - by providing fast, accurate answers to their questions. The solution delivers not only concise, relevant responses but also includes the surrounding context and direct source references, enabling analysts to validate information and make informed decisions with greater confidence and speed.
- Accelerated Insights - Instantly surface key information from massive reports like 10-Ks, 10-Qs, and earnings calls - no more endless scrolling or CTRL+F.
- Source-Cited Answers - Every answer can be grounded in and linked to the exact source text, boosting trust, traceability, and audit readiness.
- Compliance & Accuracy - Reduces the risk of hallucinations or misstatements by constraining the model to verified corporate filings and disclosures.
- Domain-Aware Without Training - Handles complex financial jargon and concepts via smart retrieval - no fine-tuning or specialized model required.
- Scalable Across Portfolios - Apply it across thousands of companies and document types, enabling analysts to compare metrics or trends efficiently at scale.
- Data Gathering
User Role(s): Data Engineer
The data engineer is responsible for gathering data from multiple sources and uploading it to datastore. The data in this use-case may come from multiple sources, mostly consisting of unstructured financial data. Elasticsearch has a large connector ecosystem that can ingest and keep in sync data from those various sources. For this example we will use an S3 connector to ingest data contained in Amazon S3 object store.
- Document Chunking + Embedding Generation
User Role(s): Data Engineer, Data Scientist, AI/MLOps
Documents are divided into smaller chunks - such as paragraphs or sections - to improve the relevance of retrieved content. Each chunk is then converted into a vector embedding using a language model and paired with metadata for traceability. Elastic has two out-of-the-box models 1) Elastic Learned Sparse Encoder Model, ELSER, which is a sparse embedding model and 2) E5 which is a dense embedding model that supports multiple languages. Elastic also has the ability to connect to third party inference services to generate embeddings. With Elastic’s Machine Learning nodes, you can define custom chunking strategies, generate embeddings, and store them directly in the vector database on the Elastic Data Node.
- Elasticsearch Vector Database Creation
User Role (s): Data Engineer, Data Scientist, AI/MLOps
Embeddings and metadata are stored in Elasticsearch which is optimized for similarity search. Elastic also stores the full text and filters like company name, document type, and fiscal year to support efficient and targeted hybrid search.
The Elastic Data Node stores the generated vector.
- Define Role Base Access Control
User Role(s): Administrators, Business Owners
Administrators apply governance through role-based access control (RBAC), ensuring users can only access the data they're authorized to see. Roles can be defined with specific permissions at the index, document, or even field level. This fine-grained control helps enforce data security, support compliance requirements, and maintain multi-tenant environments securely.
- User Initiates a Query
User Role(s): End-user (e.g: customer service agent, researcher, business analyst)
An end-user submits a natural language question (e.g., "What were Amazon’s main revenue drivers in 2023?"). The system transforms this query into an embedding and uses it to retrieve relevant document chunks after verifying the end-user has the proper privileges to access those.
A prompt is then generated and sent to the LLM. LLM responses are streamed back to the client.
- Monitoring
User Role(s): SRE (Site Reliability Engineer), Product Owner / Analyst, Business Owners
Monitoring is set up to observe metrics from infrastructure all the way to the application layer. Elastic Agents can be deployed to collect system metrics, application logs, and traces across container environments. These agents feed data into Elasticsearch, where dashboards and alerts in Kibana provide real-time insights. Elastic also enables tracking of LLM-specific metrics such as model latency, token usage, and error rates offering end-to-end visibility into both system health and user experience for RAG-based applications.
