Verity

Verity is a Gemini 3-powered forensic analysis tool that detects contradictions, verifies evidence, and reconstructs timelines across multimodal sources.

Comment 1

Inspiration

Large language models are entering courtrooms, compliance offices, and investigative workflows. But forensic analysis demands more than plausible answers. Every conclusion must be traceable, every claim backed by evidence. When I saw Gemini 3 Pro's extended reasoning capabilities, I realized we could finally build an AI system that shows its work in a domain where transparency isn't optional; it's legally required. Verity explores whether advanced AI can meet the evidentiary standards that protect justice.

What I built

Verity is a forensic evidence verification platform that analyzes multimodal evidence (documents, images, audio transcripts, and financial records) to detect contradictions, reconstruct timelines, and assess witness credibility. Unlike traditional LLM outputs, Verity exposes Gemini 3 Pro's chain-of-thought reasoning, allowing investigators to audit how conclusions were reached—not just what they are.

Key features

  1. Contradiction Detection: Cross-references claims across documents and flags logical inconsistencies with source citations
  2. Timeline Reconstruction: Builds verifiable event sequences from fragmented evidence with confidence intervals
  3. Credibility Scoring: Assesses statement reliability based on corroboration, specificity, and internal consistency
  4. Evidence Provenance: Every analytical claim links back to specific source material

How it works

Frontend: React + Vite with interactive contradiction graphs (D3.js) and animated timeline visualizations that let users drill into specific evidentiary conflicts.

Backend: Gemini 3 Pro performs multi-hop reasoning across evidence sets, comparing witness statements against physical evidence and identifying temporal impossibilities.

Demo Strategy: Due to Gemini 3 preview quota constraints, the app includes a "Golden Path" mode using pre-validated Gemini 3 Pro analyses from Google AI Studio. This ensures judges experience the model's full reasoning depth. Live uploads gracefully fall back to Gemini 2.0 Flash with transparent labeling and no hidden failures.

Challenges

Rate Limits as Design Constraint: Gemini 3 preview quotas meant I couldn't offer unlimited live analysis. Rather than hide this, I built it into the UX; the Golden Path showcases Gemini 3's capabilities on curated cases, while custom uploads demonstrate practical fallback handling. Explainability vs. Performance: Forensic users need slow, transparent reasoning, not fast black-box predictions. I designed prompts that prioritize step-by-step justification over brevity, treating latency as a feature rather than a bug.

What I learned

AI systems for high-stakes domains require different design principles. Verity taught me that:

  1. Transparency matters more than speed when decisions have consequences
  2. Graceful degradation (Gemini 3 → 2.0 Flash) can preserve trust if communicated honestly
  3. Extended reasoning models like Gemini 3 Pro unlock new use cases, but only if you design UX around auditable rather than authoritative outputs

This project convinced me that the future of AI in forensics isn't replacing human judgment; it's documenting the reasoning so humans can judge better.

What’s Next for Verity

Verity is an exploration of what explainable, audit‑ready AI can look like in forensic and legal workflows. Future development would focus on:

  1. Live Gemini 3 Scaling
    Integrating full Gemini 3 Pro access as preview limits relax, enabling real‑time analysis for complex, user-submitted evidence without fallback constraints.

  2. Chain‑of‑Custody & Audit Logs
    Adding immutable evidence logs and exportable reasoning trails to support legal review, compliance audits, and courtroom use.

  3. Human‑in‑the‑Loop Review
    Allowing investigators to annotate, challenge, and approve AI reasoning steps, positioning Verity as a decision‑support system, not a decision‑maker.

  4. Expanded Multimodal Forensics
    Deeper analysis of audio inconsistencies, image manipulation detection, and metadata‑level verification across more evidence types.

  5. Enterprise & Regulatory Readiness
    Role‑based access control, case isolation, and policy‑driven reasoning constraints for regulated environments.

Verity’s long-term goal is not to replace forensic experts, but to raise the standard of AI-assisted investigations by making reasoning transparent, traceable, and accountable.

Share this project:

Updates