Skip to content

Latest commit

 

History

History
58 lines (41 loc) · 2.74 KB

File metadata and controls

58 lines (41 loc) · 2.74 KB

📂 Submission Media Index

This folder contains the visual evidence for Saarthi AI's submission to the Convolve 4.0 Hackathon (Qdrant Track). Each file below has been captured from the live, running application to demonstrate specific technical claims.

Suggested order for review:
Video → Qdrant Proof → Memory Before/After → Eligibility Block → UI Highlights


🎥 Primary Demonstration

  • core_walkthrough_demo_*.webp
    • What it is: A ~2-minute end-to-end walkthrough of the user journey.
    • Flow: Landing Page → Semantic Search → Hybrid RRF Results → Explainable Eligibility → Agentic Fallback.
    • Note: This video serves as the primary visual reference for the system's fluidity and speed.

🕵️ Forensic Proofs (Technical Validation)

These screenshots validate the backend logic and invisible system architecture. (Implementation details: Technical Report Sections 6, 8, 9, 14)

  • qdrant_collections_proof_*.png

    • Claim: "Dual-Collection Architecture"
    • Evidence: Shows the mas_scholarships (Knowledge) and mas_interactions (Long-term Memory) collections running side-by-side in the local Qdrant instance.
  • memory_before_ranking_*.png & memory_after_ranking_*.png

    • Claim: "Closed-Loop Retrieval with Persistent Vector Memory"
    • Evidence: A "Before & After" comparison showing how a user interaction (clicking a result) instantly re-ranks future search results for that specific user, boosting relevant items.
  • eligibility_blocked_proof_*.png

    • Claim: "Deterministic Safety Layer"
    • Evidence: Shows a "Rich Student" profile intentionally being blocked from a means-based scholarship. The red flags verify that the deterministic rules engine overrides vector similarity.

💎 UI & Feature Highlights

These images demonstrate the "Cyber-Glass" design language and observability features.

  • landing_hero_glass_*.png

    • Feature: Premium "First Impression" UI with glassmorphism and motion design.
  • search_results_radar_*.png

    • Feature: The "Hybrid Search" interface showing match scores, verification badges, and the "Explainable Fit" radar chart.
  • scholarship_details_view_*.png

    • Feature: Deep-dive view showing the 100-point eligibility breakdown and transparent scoring logic.
  • agent_memory_log_*.png

    • Feature: The "Glass-Box" Agent stream, revealing the system's internal reasoning steps to the user in real-time.

Authenticity Note
All assets were captured from a live local deployment.
No mockups, no static UI, no staged data.
Qdrant ran via Docker with dual collections active. Verified by Saarthi AI