FitMatch Reasoning Coach

Inspiration

Online shopping today is driven by speed and influence. We see a TikTok, a YouTube haul, or an influencer post — and within seconds we’re convinced.

What it does

FitMatch Reasoning Coach helps users think critically before buying a product online.

Users paste a product URL they’re considering. The app then:

Extracts product metadata

Pulls YouTube review evidence

Classifies reviews (neutral vs sponsored)

Detects persuasion signals (scarcity, hype, authority, social proof)

Breaks the claim into testable sub-claims

Highlights contradictions and missing evidence

Guides users through structured reflection questions

Calibrates confidence before and after reviewing evidence

Instead of telling users what to buy, we:

Transform hype into evidence and let the user decide.

How we built it

FitMatch is built using:

Next.js + TypeScript for the frontend and routing

API routes for evidence fetching and analysis

YouTube Data API to pull review content

Mistral LLM (optional layer) for structured reasoning and verdict generation

Heuristic fallback logic when APIs are unavailable

LocalStorage for session history and reflection tracking

We structured the system around a reasoning pipeline:

Claim → Evidence → Contradictions → Bias Map → Reflection → Confidence → Verdict

Challenges we ran into

Designing the right number of questions Too many questions overwhelm users. Too few reduce depth. Finding the balance between usability and rigor was difficult.

Scraping and sourcing evidence Accessing reliable review data while respecting API limits and structure required careful design.

Distinguishing persuasion from evidence Influencer content often mixes genuine opinion with sponsored promotion. Separating signal from noise required layered logic.

Making it feel intelligent, not generic We wanted structured reasoning — not vague AI summaries.

Accomplishments that we're proud of

Building a full claim-to-verdict reasoning pipeline

Classifying sponsored vs neutral reviews

Designing a Bias Map and Claim Stress Test

Implementing confidence calibration before and after evidence review

Creating a prototype that shifts decision power back to the user

Most importantly:

We built a system that encourages thinking, not impulse.

What we learned

Influence is subtle — especially in short-form video.

Most reviews are overwhelmingly positive, which creates perception bias.

Users need structure to think critically.

AI should guide reasoning, not replace it.

Clear visual presentation matters as much as backend intelligence.

What's next for FitMatch Reasoning Coach

Next steps include:

Integrating TikTok sponsored video analysis

Deeper YouTube transcript parsing

Detecting affiliate links and discount code patterns

Cross-platform evidence aggregation (Reddit, blogs, forums)

Personalization based on user values and budget

Real-time influence distortion scoring

Mobile-first experience

Long-term vision:

FitMatch becomes a real-time Influence Intelligence layer for online shopping.

Built With

• api
• data
• mistral-ai-api
• next.js
• react
• tailwind-css
• typescript
• youtube