PM

Prince Malani

AI Specialist & Data Scientist
INITIATING ENVIRONMENT... [INIT]
← Return to Projects Portfolio

UNLEASH LLM Innovation Challenge

Built the official platform for the Unleash LLM Innovation Challenge, a national hackathon under the YuvAI Initiative, organized with Meta in collaboration with IndiaAI, AICTE, and implemented by 1M1B. The portal manages the complete hackathon lifecycle across 7 themed categories, guiding participants through a structured two-phase journey: team registration with unique Team ID generation, idea submission via one-page PDF reports, and prototype submission through walkthrough videos and GitHub repository links for shortlisted teams.

Author: Prince Malani | Technologies: React, Python, FastAPI, JavaScript, MongoDB

Project Overview & Architecture

# Unleash LLM Innovation Challenge Management Platform The **Unleash LLM Innovation Challenge Platform** is an enterprise-grade competition management SaaS engineered to power high-concurrency, nationwide hackathons and technology innovation challenges. Built for the national **YuvAI Initiative**—a flagship partnership between **Meta**, **IndiaAI (Digital India Corporation / MeitY)**, **AICTE**, and **1M1B (One Million for One Billion)**—the platform facilitates capacity building and open-source LLM adoption across thousands of young developers and engineering students in India. The application governs the end-to-end lifecycle of multi-stage technological challenges across key impact tracks: *Inclusive Health, Climate & Environment, Education & Skills, Agriculture & Rural Development, Smart Cities, Financial Inclusion,* and *Open Innovation*. --- ## Problem Statement Managing national-scale technical competitions across hundreds of universities presents significant structural, operational, and data-integrity challenges. Traditional methods often fall short when dealing with high user volume and complex workflows. * **Fragmented Evaluation & Submission Workflows:** Traditional hackathons rely on disconnected tools (e.g., forms, spreadsheets, disparate cloud folders), leading to missing project files, mismatched team credentials, and high operational latency. * **High-Bandwidth Multi-Format Asset Intake:** Managing dual submission phases—Phase 1 structured PDF idea reports and Phase 2 prototype video pitches and code repositories—strains server storage and risks payload timeouts without robust file handling. * **Manual Phase Progression Bottlenecks:** Opening and closing submission windows across thousands of users typically requires manual deployment cycles or manual database edits, risking downtime and race conditions. * **Administrative Overhead & Evaluation Friction:** Program administrators and jury panels struggle to aggregate multi-thousand team entries, verify validation rules, generate consolidated evaluation sheets, and download gigabytes of submission assets without crashing servers. * **Communication Latency:** A lack of automated transactional confirmation engines leaves participants uncertain regarding submission validity, team ID assignment, and stage qualification status. --- ## Solution Architecture The platform is designed around a modular, decoupled SaaS architecture prioritizing asynchronous processing, data isolation, and non-blocking I/O operations for heavy file ingestion and retrieval. It bridges the operational divide between high-volume public participation and strict administrative governance. ### Core Architectural Components * **State-Driven Workflow Engine:** Enforces strict state transitions across registration, Phase 1 (Idea PDF submission), Phase 2 (Prototype & Video Pitch submission), and Jury Review stages. * **Centralized Configuration & Feature-Flag Registry:** Controls dynamic visibility, submission window toggles, and form routing dynamically from an administrative control dashboard. * **Decoupled Asset Ingestion Pipeline:** Validates file signatures, MIME types, and file size constraints on intake before persisting metadata to the database and routing raw assets to secure storage buckets. * **Streaming Archival Engine:** Utilizes memory-efficient streaming workers to compress gigabytes of PDF documents, metadata matrices, and media links into single ZIP bundles on-the-fly without overloading memory. --- ## Technical Execution The development and deployment of this platform required robust engineering practices to ensure scale and reliability. * **Multi-Stage Pipeline & Dynamic Team Identity:** Developed an automated identity generation pipeline that assigns collision-free, track-specific Unique Team IDs upon initial registration. Built multi-tier submission forms with client- and server-side schema validations for team member details, LLM usage briefs, and one-page PDF reports. * **Real-Time Admin Feature Flagging:** Architected a runtime feature-flagging system enabling program directors to toggle registration gates, prototype submission portals, and result visibility instantly without codebase redeployment. Implemented role-based access control (RBAC) separating administrative governance from participant portals. * **Asynchronous Transactional Email Service:** Integrated a secure SMTP transactional mail pipeline utilizing responsive, sanitized HTML templates. Configured automated dispatch of submission receipts, assigned Team IDs, deadline reminders, and stage-advancement notices with delivery retry logic. * **Memory-Efficient Batch Export & Archival:** Built a high-performance streaming export pipeline capable of compiling multi-sheet Excel workbooks containing detailed team metadata and submission links. Integrated dynamic ZIP streaming using chunked data transfer protocols to allow administrators to download gigabyte-scale asset packages without OOM exceptions. * **Storage Integrity & Lifecycle Automation:** Configured automated file purging and cascading delete constraints to prevent orphaned assets in local/cloud storage when submissions are re-uploaded, rejected, or modified. --- ## Platform Workflow & Participant Journey The platform provides a seamless self-service experience guiding student teams from ideation to final showcase. 1. **Phase 1: Registration & Idea Submission:** Teams register online, receiving automated Team IDs. They submit project ideas detailing their tech stack and open-source LLM usage via validated PDF reports. 2. **Phase 2: Prototype Build & Scrutiny:** Shortlisted teams use a gated gateway to submit prototype video pitches, minor code demonstrations, and repository links securely. 3. **Phase 3: Administrative Filtering & Export:** Jury and administrators utilize one-click chunked ZIP downloads and comprehensive Excel data exports to review entries without server lag. 4. **Phase 4: Finale Showcase:** The top selected teams move forward to the virtual finale, managed seamlessly through the platform's state-driven dashboard. --- ## Results & Impact * **High-Throughput Operations:** Successfully powered the operational backbone for the national YuvAI Initiative, coordinating registrations across all 7 impact tracks smoothly. * **Zero Operational Downtime:** Sustained peak submission surges near deadline cutoffs without server memory exhaustion or dropped submissions. * **Reduced Administrative Overhead:** Decreased manual administration and file collation time by over 90% through automated ZIP packaging and multi-sheet Excel exports. * **Enhanced Participant Experience:** Achieved 100% automated confirmation receipt delivery, eliminating support query backlogs regarding submission status. * **Scalable Open Innovation:** Provided a structured, resilient evaluation funnel mapping nationwide registrations down to the Top 3 National Winners. --- ## Key Learnings * **Asynchronous File Handling at Scale:** Managing high-volume file intake during competitive deadlines highlighted the absolute necessity of streaming file archives and decoupling heavy file processing from the primary request-response loop. * **Runtime Config Over Hardcoded Schedules:** Implementing dynamic runtime feature flags proved far more resilient than time-based cron scheduling, giving organizers flexibility to extend deadlines without risky redeployments. * **Defensive Data Schemas:** Designing data schemas that anticipate edge cases (duplicate emails, track switching, member dropouts) in the early stages prevented major data anomalies downstream during jury evaluation. * **System Observability:** Centralized logging of transactional email dispatches and upload pipelines proved essential for resolving participant inquiries quickly during high-stakes national events.

View All Projects | Discuss this Project