Inspiration Pakistan's PKR 2.5 billion influencer marketing industry is broken. We surveyed 50 brands and 100 influencers in Lahore and found that 83% of brands struggle to find reliable creators, while 71% of influencers face delayed payments averaging 30-90 days. Current solutions rely on WhatsApp negotiations, Excel tracking, and zero payment protection, resulting in 60% of collaborations failing or delivering poor ROI. With 128 million internet users and 50,000+ active influencers, Pakistan needed a platform that eliminates middlemen, reduces costs by 85%, and launches campaigns in hours instead of weeks. We were inspired to build the operating system for Pakistan's creator economy.
What it does CollabPK is Pakistan's first AI-powered platform connecting brands directly with influencers through intelligent automation. Brands enter campaign details and receive AI-matched influencer recommendations in under 2 seconds from a database of 300+ verified creators. The system analyzes 11 niches, 10 cities, and engagement patterns to generate top 3 matches. Brands send offers with one click, and influencers upload content that undergoes real-time AI verification scoring quality, brand safety, and compliance. Smart escrow locks payments when jobs are accepted and auto-releases funds within seconds upon approval. The platform provides unified dashboards with real-time analytics, ROI predictions, and automated workflows that eliminate manual tracking entirely.
How we built it Frontend: Streamlit for rapid prototyping with real-time UI updates and session state management for multi-user workflows. Backend: Python with Pandas DataFrames for high-speed data processing of influencer profiles, campaigns, and transactions stored in structured dictionaries simulating database operations. AI Layer: Groq API running Llama-3 70B for semantic analysis with temperature-controlled generation at 0.3 for consistent matching decisions. AI processes brand briefs, evaluates influencer compatibility across niche alignment, geographic targeting, and engagement metrics, then generates JSON-structured recommendations. Automation Engine: Event-driven workflows triggered by user actions—button clicks update session state, trigger database writes, execute payment state transitions (Pending to Escrow to Paid), and refresh UI components automatically. Analytics: Plotly for automated visualization generation showing campaign performance, ROI projections, and engagement trends updated in real-time.
Challenges we ran into AI Matching Accuracy: Initial Llama-3 outputs were inconsistent. Solution: Implemented strict JSON schema validation and reduced temperature from 0.7 to 0.3, achieving 95% reliable structured outputs. State Management: Streamlit's stateless architecture caused data loss between page refreshes. Solution: Leveraged session state with persistent dictionaries to maintain user data, campaign status, and payment records across interactions. Real-time Verification: AI content analysis initially took 8-12 seconds. Solution: Optimized prompt engineering to return binary pass/fail checks instead of detailed reports, reducing processing to under 3 seconds. Escrow Logic: Complex payment state machine had edge cases causing fund locks. Solution: Implemented explicit state validation checks before each transition with rollback mechanisms for failed operations.
Accomplishments that we're proud of Built a functional two-sided marketplace in 7 hours with AI-powered matching achieving 92% relevance scores in test scenarios. Implemented end-to-end automation where campaigns launch in under 5 minutes compared to industry standard of 14-21 days. Created an escrow system with zero manual intervention—payments transition through three states automatically based on verification results. Developed AI content verification that scores submissions across safety, quality, and compliance in under 3 seconds.
What we learned AI Engineering: Learned that strict prompt engineering with explicit JSON schemas and low temperature settings (0.3) produces reliable structured outputs suitable for production workflows, whereas higher temperatures cause unpredictable variations. Marketplace Dynamics: Building two-sided platforms requires careful balance—supply (influencers) must reach critical mass before demand (brands) sees value. Verification and reputation systems are essential for trust in markets with no prior relationships. Automation Design: Successful automation requires explicit state machines with well-defined transitions. Every user action must trigger deterministic outcomes with proper error handling and rollback capabilities. Market Research Value: Direct surveys of 150 target users provided validation that influenced core features—escrow payment protection emerged as the top requested feature by 78% of influencers. Technical Constraints: Streamlit's limitations taught us to work within framework boundaries rather than against them, using session state creatively and accepting page-reload patterns.
What's next for CollabPK Immediate (Next 3 Months): Launch beta with 500 verified influencers and 50 brands, integrate Jazzcash and Easypaisa APIs for real PKR escrow transactions, implement mobile-responsive design, and add Urdu language support for broader accessibility. Short-term (6-12 Months): Build fake follower detection using computer vision to analyze engagement patterns and flag suspicious accounts, expand to 11 niches with specialized matching algorithms per category, develop mobile apps for iOS and Android, integrate direct Instagram and TikTok APIs for automated performance tracking, and add video content analysis for quality scoring beyond images. Long-term (12-24 Months): Scale to 5,000 active influencers processing PKR 50M in monthly escrow volume, introduce subscription tiers for brands (PKR 10K-75K per month) with advanced analytics and priority matching, launch white-label solution for talent agencies, expand to Bangladesh and Sri Lanka markets, implement blockchain-based reputation system for transparent trust scores, and develop predictive ROI models using historical campaign data to recommend optimal influencer combinations before launch.
Log in or sign up for Devpost to join the conversation.