Smart Document Search for a B2B Accounting Platform Built in One Week with Dynamiq

Summary

A B2B accounting SaaS company partnered with Dynamiq to help users quickly find legal and financial documents using natural language. Dynamiq’s low-code platform — powered by Anthropic Claude 3.7 Sonnet on AWS Bedrock, Amazon Titan Text Embeddings v2, and an OpenSearch Vector DB — let the team spin up a production-grade RAG search feature in under a week.

The result was a production-ready AI feature that can understand queries like "Show me all NDA contracts signed last month," or "What contracts are set to expire in the next month?" directly inside the client’s app.

The Challenge

The client’s B2B accounting platform handles large volumes of unstructured documents: contracts, invoices, presentations, and more. Users often needed to find specific information buried deep within these files, but the platform’s traditional search features weren’t flexible or accurate enough.

The client also wanted to let users ask questions using natural language, for example:

• Which contracts expire next month?
• Find invoices sent to [Client Name] in March.
• Show me invoices from Q1 with overdue payments.
• List all expired contracts that haven’t been renewed.
• What policies were updated in the last 60 days?
• Find presentations related to our Series B funding.
• Which documents contain references to “cash flow forecast”?

Additionally, building a semantic search capability in-house would have required significant time and resource investment.

Overall, the client needed a solution that would:

• Support both semantic and keyword-based search
• Handle complex document types (PDFs, Word, PowerPoint, images)
• Scale easily with minimal infrastructure overhead
• Get to production quickly

Why Dynamiq

Following a recommendation from a partner, the client reached out to Dynamiq. We offered a way to build and deploy a production-ready intelligent search system without the need for a dedicated ML or DevOps team in days instead of months.

Here is what makes Dynamiq stand out:

• A low-code builder for configuring workflows, pipelines, and integrations
• Out-of-the-box support for retrieval-augmented generation (RAG)
• Seamless integration with Amazon OpenSearch as a Vector Database for lightning-fast vector and hybrid search.
• Flexible deployment options for secure, enterprise-grade environments.
• One-click connections to Anthropic Claude 3.7 Sonnet and Titan Text Embeddings v2 via AWS Bedrock — no custom pipelines or DevOps overhead.

The Solution

Dynamiq allowed the client’s in-house team to build and deploy a RAG-based search system that now serves as the intelligent backend of the client’s platform document search.

Here's how it works:

1. Preprocessing. Various unstructured documents (PDFs, DOCX, PPTX, images) are ingested and embedded.
2. Vectorization. Stored in Amazon OpenSearch for sub-second nearest-neighbor look-ups.
3. Retrieval. Dynamiq orchestrates a hybrid pipeline: Titan Text Embeddings v2 vectors + keyword scoring → Anthropic Claude 3.7 Sonnet (via AWS Bedrock) synthesizes answers with citations.
4. Deployment. Thanks to Dynamiq’s low-code interface, the search functionality was built by a single forward deployed engineer in less than a week.

Results

• Fully functional natural language search
• No additional infrastructure or custom ML models required
• Reduced development time and accelerated time-to-value

What’s Next

The client is continuing to expand its AI capabilities with Dynamiq, focusing on:

• Document summarization
• Structured data extraction (e.g., contract terms, payment details)
• Invoice parsing and classification