Project overview
The problem we solved
Brands, agencies, and SaaS teams need YouTube creators who drive revenue — not vanity metrics. Evaluating a channel means watching dozens of videos, reading thousands of comments, and checking sponsor history. Most teams don't have the analysts to do it at scale, and expensive agencies deliver outdated shortlists weeks later.
The challenge
What teams were stuck with
Marketing managers and partnership leads were stuck between spreadsheets and gut feel — with no shared framework for brand safety, ROI, or repeat sponsorship signals.
- Manual YouTube research took weeks per campaign and couldn't keep pace with creator churn
- Subscriber counts hid dead audiences, low buying intent, and one-off sponsor deals
- No scalable way to search across hundreds of niches, languages, and audience profiles
- Outreach stayed generic because teams never had time to personalize at volume
- Partnership decisions were subjective — no shared language for marketing, legal, or leadership
Our solution
What we built
Vyrex ingests channel metadata, video content, comments, and transcripts — then produces partnership scores, brand-safety assessments, and clear TEST, WAIT, or SKIP recommendations.
- Automated discovery across 250,000+ keyword variations
- Deep per-channel analysis: commercial probability, brand repetition, momentum, and risk
- Natural language search across the full creator database via hybrid semantic + structured queries
- AI-generated outreach sequences tailored to each creator's content and audience
- Transparent decision matrix for brand awareness, direct sales, SaaS, B2B, and long-term partnerships
Technology
Built with enterprise-grade technology
Goals and objectives
What the platform needed to deliver
01
Automate creator discovery
Continuously surface new channels across business, consumer, and international niches without manual keyword hunting.
02
Replace guesswork with intelligence
Score creators on buying intent, sponsor repeat rate, momentum, and brand safety — not subscriber count.
03
Search like you think
Let users describe ideal partners in natural language and return ranked, filterable results from a vector-indexed database.
04
Accelerate outreach
Generate personalized multi-touch email sequences from real channel analysis, not generic templates.
05
Scale analysis cost-effectively
Route high-volume tasks to fast models and reserve premium reasoning for final partnership decisions.
06
Build a reusable database
Every analyzed channel becomes searchable, exportable, and comparable across campaigns.
07
Pay-per-insight monetization
Unlock full reports on demand with credits and Stripe — preview before you commit.
Solution in action
See the platform in action
From discovery to outreach — how Vyrex transforms partnership workflows.
01
Platform overview
Vyrex positions brands to find creators who actually drive sales — powered by 100K+ analyzed influencers with partnership scores, not vanity metrics.
02
Natural language search
Teams describe ideal partners in plain English. GPT parses filters; pgvector runs hybrid semantic search across readiness, engagement, growth, niche, and geography.
03
Ranked search results
Thousands of matching channels returned with partnership scores, niche tags, audience demographics, brand-safety signals, and one-click export to CSV.
04
Partnership scoring & brand fit
Each profile includes a decision matrix — commercial probability, brand repetition, audience snapshot, and a clear TEST, WAIT, or SKIP verdict teams can share internally.
05
Deep analysis & recommendations
Audience pain points, risk profile, key strengths, and partnership recommendations — grounded in real viewer behavior from comments and transcripts, not channel bios.
Platform architecture
How it all works together
1
Discovery engine
Celery workers run scheduled discovery across niche keyword libraries, deduplicating channels and caching stats to protect API quota.
2
Parallel video analysis
Each video analyzed concurrently — comments fetched, transcripts via AssemblyAI, eight AI dimensions evaluated in parallel.
3
Tiered AI model router
Gemini Flash for high-volume tasks, GPT-4o-mini for reasoning, GPT-4 for critical unlock decisions — optimizing cost without sacrificing quality.
4
Partnership intelligence layer
Proprietary scoring combines commercial probability, sponsor repetition, 90-day growth, engagement ratios, and brand-safety into a readiness score.
5
pgvector search index
Every analysis embedded with 40+ structured fields — subscribers, niche, pain severity, ROI potential, promoted brands — enabling hybrid search at scale.
6
Outreach & monetization
Stripe-backed credits, background email generation, CSV export, and API endpoints connect analysis to revenue action.
Results
Value and impact delivered
Measurable improvements across the partnership workflow.
Research timeline: weeks → hours
Discovery, analysis, and ranked shortlists run as background jobs — a fraction of the time manual digging took.
False positives eliminated
Engagement checks, brand repetition, and buying-intent scoring surface creators who convert — not inflated subscriber counts.
Search the entire database
Natural language queries replace static spreadsheets. Filter by growth, niche, promoted brands, score, and geography without SQL.
Outreach at scale, still personal
AI email sequences reference each creator's actual content and audience — researched, not spammed.
Decisions teams can defend
Every profile includes a decision matrix and explicit risk summary — shared language for marketing, legal, and leadership.
Platform built to scale
Modular architecture supports 500K+ channel targets, distributed workers, vector indexing, and continuous niche expansion.