NOVA RISK ANALYST

Nova Risk Analyst is a state-of-the-art financial fraud detection platform that replaces hours of manual document review with seconds of AI-powered reasoning. #AmazonNova

Comment

Inspiration

Financial compliance is currently a battle of manual labor versus high-speed digital crime. We saw compliance officers spending hundreds of hours manually squinting at low-quality bank statement scans and complex spreadsheets to find "needles in haystacks" like money laundering or structuring. We were inspired to build a solution that doesn't just process data, but understands the financial narrative. By combining the deterministic precision of rules with the reasoning power of Amazon Nova 2 Lite, we wanted to democratize elite-level financial oversight for small and mid-sized institutions.

What it does

Nova Risk Analyst is a multi-layered financial document intelligence platform. It allows analysts to upload bank statements (as images or PDFs) and transaction logs (CSVs). The system then runs a 3-stage pipeline:

  1. Multimodal Extraction: Amazon Textract pulls structured data from the visual layout.
  2. Deterministic Shield: A Python rule engine flags 12 critical violations like velocity spikes and round-number patterns.
  3. AI Reasoning Layer: Amazon Nova 2 Lite analyzes the extraction and the rules to separate true threats from benign anomalies, providing a 0-100 risk score and a natural-language executive summary. It also features an Interactive Auditor chat where analysts can ask follow-up questions directly about the document's context.

How we built it

We built Nova Risk Analyst using a modern cloud-native stack:

  • Backend: FastAPI for high-performance async processing.
  • AI/ML: AWS Bedrock serving the Amazon Nova 2 Lite model. We leveraged its multimodal capabilities to reason over both text and image data simultaneously.
  • OCR: Amazon Textract for high-fidelity table and form extraction.
  • Frontend: A premium, dark-mode single-page application built with Vanilla JS, CSS3, and Chart.js for data visualization.
  • Architecture: A secure 3-stage pipeline that ensures "Human-in-the-Loop" transparency with a full AI reasoning audit trail.

Challenges we ran into

One of the primary challenges was handling the "messiness" of real-world financial documents—blurry scans, complex nested tables, and varying formats. We solved this by using Amazon Textract’s specialized table features instead of generic OCR. Another challenge was Bedrock's sensitive word guardrails; we implemented a "compliance sanitizer" that translates high-trigger terms into neutral audit language (e.g., "money laundering" to "irregular fund movement") to ensure seamless processing while maintaining investigative accuracy.

Accomplishments that we're proud of

We are incredibly proud of the Interactive Auditor feature. It transforms the compliance report from a static PDF into a living investigation partner. The ability for an analyst to ask "Why was this specific wire transfer flagged?" and get a cited, evidence-backed answer from Nova 2 Lite is a game-changer for transparency. We are also proud of our "Demo Mode," which allows users to experience the full power of the dashboard instantly without needing their own API keys upfront.

What we learned

Building with the Amazon Nova family taught us a lot about the balance between reasoning speed and multimodal accuracy. We learned that Nova 2 Lite is exceptionally capable of identifying complex "layering" patterns that simple mathematical rules miss. We also deepened our understanding of the AWS Bedrock ecosystem and how to build secure, auditable AI pipelines that respect data privacy—a non-negotiable requirement in the financial sector.

What's next for NOVA RISK ANALYST

The roadmap for Nova Risk Analyst is ambitious:

  1. Multi-Document Sequencing: Allowing analysts to upload multiple months of statements to detect long-term historical trends.
  2. Nova Act Integration: Automating the generation of Suspicious Activity Reports (SARs) directly into regulatory portals.
  3. Enterprise Connectors: Building native integrations for common accounting platforms and core banking systems.
  4. multimodal 2.0: Using Nova's video understanding capabilities to analyze KYC (Know Your Customer) video verification sessions for liveness and fraud.

Built With

Share this project:

Updates