Contrast AI

The description states that Contrast AI:

• Accepts YouTube links or voice notes via browser
• Transcribes content using OpenAI's audio API
• Extracts verifiable claims from the transcription
• Gathers public evidence with OpenAI Web Search
• Cross-checks claims against multiple independent sources
• Produces a report including:
  • An alert score (0–100) with uncertainty range and coverage
  • Claim breakdown: supported, contradicted, missing context, unverified
  • Persuasion analysis detecting 13 manipulation techniques
  • Side-by-side contrasts between content and evidence
• Operates under two principles:
  • Evaluates content, not person
  • When evidence is insufficient, verdict is "inconclusive"
• Is bilingual (English/Spanish)

The system uses a multi-agent pipeline powered by GPT-5.6 via OpenAI Agents SDK. It includes:

• An orchestrator identifying claims
• A discourse analyst mapping persuasion tactics
• Research agents gathering evidence in parallel
• GPT-5.6 adjudicating claims based on retrieved evidence (cannot search independently)
• Voice notes are chunked with ffmpeg and transcribed before entering pipeline

The web app is built with Next.js + Mantine, and the backend uses Supabase-backed workers deployed as Railway services.

Inference The product appears to be a proof-of-concept or prototype built in a short timeframe (a hackathon), not a commercial-grade solution. It has no evidence of production deployment or user base beyond its own description.

The description states that Contrast AI was inspired by the need for everyday fact-checking, which is currently slow and manual. The authors claim:

• They wanted anyone to be able to check what a video claims before acting on it
• Fact-checking exists but is reserved for headline news — not daily content shaping decisions
• Their tool aims to democratize access to verification

They also state that the system evaluates content, not people, and refuses to manufacture certainty when evidence is lacking.

Inference The positioning is framed as a tool for personal empowerment in an information-rich environment where manipulation is common. However, there is no evidence of prior market research or user feedback beyond the hackathon context.

The description states that Contrast AI targets individuals who make financial, personal, or health decisions based on content like videos and voice messages.

It does not specify:

• Whether it targets creators, media organizations, educators, or general consumers
• If there are distinct personas within this group
• Any segmentation strategy or customer journey

Inference The ICP is likely broad — anyone who consumes video/audio content and wants to verify claims. But no evidence of specific targeting or user research exists.

The description does not state:

• How Contrast AI intends to monetize
• If it plans to charge users, offer freemium tiers, or sell to enterprises
• Whether pricing exists or is planned
• Any revenue model or monetization strategy

Inference No business model or pricing evidence is provided. The project appears to be a prototype without commercial intent.

The description states:

• Built with Next.js, Supabase, TypeScript
• Uses OpenAI’s audio API and Web Search
• Multi-agent pipeline powered by GPT-5.6 (OpenAI Agents SDK)
• Voice notes are processed via ffmpeg + OpenAI audio API
• System avoids hallucinations through deterministic code for scoring and linking
• Input validation, private storage, job queuing with retries and health checks
• Shared Zod contracts between components
• UI is bilingual (English/Spanish)

Inference The technical stack suggests a modern, scalable architecture built in a short timeframe. However, there's no evidence of production deployment or performance metrics.

The description states:

• Built during a 1-week hackathon
• No prior version or commercial history is mentioned
• No revenue, customers, or adoption data are provided
• The team size is 3 people
• The system was built in collaboration with Codex

Inference There is no evidence of traction, revenue, or customer adoption. This is a prototype, not a mature product.

The description does not mention:

• Direct competitors
• Existing tools for content verification or fact-checking
• Market size or competitive landscape
• Any differentiation strategy beyond the hackathon context

Inference No competitive analysis or positioning relative to existing players is evident. The project appears to be in a nascent stage with no known market presence.

The description states:

• LLMs are kept honest through deterministic code and evidence-based scoring
• Risk of attribution when creator identity is ambiguous — resolved by discarding public context
• Challenges include transcript quality, AI hallucinations, retry logic, and input validation

Inference The project lacks commercial viability indicators and may be vulnerable to technical or ethical risks if scaled without further development.

1. What is the intended business model for Contrast AI?
2. Has there been any user testing beyond the hackathon?
3. How does Contrast AI plan to scale its research agents and evidence gathering?
4. Are there plans to integrate with existing fact-checking or media verification platforms?
5. What are the legal and ethical implications of flagging persuasion techniques in content?
6. How will Contrast AI handle edge cases like deepfakes, misinformation, or biased sources?
7. Is there a roadmap for monetization or commercial partnerships?