VeriWeave - AI-Powered Forensic Analysis

AI-powered multimodal forensic analysis platform using Gemini 3 to detect document and media fraud through batch cross-file reasoning.

Comment

Inspiration

Document and media fraud is a growing global problem affecting finance, healthcare, insurance, and legal systems. Existing verification tools analyze files individually, making it easy for coordinated fraud across multiple documents to go undetected.

We were inspired to solve a critical gap: What if AI could analyze multiple files together and reason across them like a forensic investigator?

With Gemini 3’s multimodal reasoning capabilities, we saw an opportunity to build a platform that could detect cross-document inconsistencies and provide unified authenticity insights — leading to the creation of VeriWeave.

What it does

VeriWeave is an AI-powered forensic analysis platform that verifies the authenticity of documents, images, and videos. Its core innovation is Batch Multimodal Analysis — sending multiple files to Gemini 3 in a single API call to perform cross-file reasoning.

The platform provides:

Authenticity score (0–100%) Risk level assessment Forensic signal detection Logical consistency checks Synthetic media detection Professional PDF forensic reports

It supports: Single file verification Batch cross-file analysis Multi-case investigation workflows

How we built it

We built VeriWeave as a full-stack AI system:

Frontend: React + TypeScript Tailwind CSS Vite

Backend: Node.js + Express Multer for file uploads AI Integration Google Gemini 3 Pro Preview (gemini-3-pro-preview) Multimodal prompt engineering Structured JSON outputs Database MongoDB Atlas for analysis storage

Deployment: Vercel (frontend) Render (backend)

Batch processing is powered by sending multiple files in one request: const contentArray = [prompt, file1, file2, file3]; await model.generateContent(contentArray);

Challenges we ran into

Designing prompts for cross-file reasoning instead of single analysis Handling multiple file formats (images, PDFs, videos) together Structuring Gemini outputs into usable forensic reports Managing large file uploads with low latency Ensuring consistent authenticity scoring across cases Each challenge required experimentation in prompt design, backend optimization, and multimodal preprocessing.

Accomplishments that we're proud of

Built a working batch multimodal forensic engine Enabled cross-document fraud detection Generated structured forensic reports via Gemini 3 Deployed a fully functional public platform Implemented three analysis modes Delivered real-time authenticity scoring Most importantly, we demonstrated a real-world use case of Gemini 3’s advanced reasoning capabilities.

What we learned

Multimodal AI can replicate investigative reasoning Context depth improves fraud detection accuracy Prompt engineering is critical for structured outputs Cross-file analysis is far more powerful than isolated verification Gemini 3 excels at correlating evidence across formats This project expanded our understanding of applied AI in high-impact domains.

What's next for VeriWeave - AI-Powered Forensic Analysis

We plan to evolve VeriWeave into an enterprise-grade forensic platform with: Blockchain-backed authenticity logs Real-time API integrations for banks & insurers Deepfake video detection expansion Government document verification pipelines Automated audit trail analysis Foundation & investigation agency partnerships Our long-term vision is to build a global AI authenticity infrastructure powered by Gemini-class multimodal reasoning.

About the gemini model

VeriWeave showcases Gemini 3 Pro Preview's breakthrough multimodal reasoning through batch cross-file analysis - a capability that sets it apart from traditional single-file verification tools. We send multiple documents (images, videos, PDFs) to Gemini 3 in a single API call, enabling the model to perform unified forensic reasoning across the entire document set.

Key Gemini 3 Features Used:

  1. Batch Multimodal Reasoning (Core Innovation): Multiple files analyzed together in one request, allowing Gemini 3 to detect cross-file inconsistencies, correlate information, and provide unified authenticity conclusions. This demonstrates Gemini 3's advanced reasoning across multiple documents simultaneously.

  2. Multimodal Processing: Gemini 3 processes images, videos, and PDFs together, extracting visual, textual, and structural information in parallel.

  3. Advanced Prompt Engineering: Detailed forensic prompts guide Gemini 3 through six analysis dimensions: multimodal matching, document forensics, visual artifacts, logical consistency, synthetic signs, and shadow/perspective analysis.

  4. Structured JSON Output: Gemini 3 returns comprehensive analysis including authenticity scores (0-100), risk levels, category-specific confidence scores, technical forensic signals, and detailed reasoning - all in structured JSON format.

Why This Matters: Traditional tools analyze files in isolation. VeriWeave's batch analysis enables Gemini 3 to detect fraud patterns across document chains (e.g., invoice + receipt + bank statement), making it powerful for real-world fraud detection where inconsistencies only appear when documents are analyzed together.

This application demonstrates Gemini 3's superior reasoning in a critical real-world context: preventing billions in document fraud through advanced multimodal AI.

Built With

Share this project:

Updates