Grant Application: The On-Chain Hero - Param Singh

applications/on_chain_hero.md

@@ -0,0 +1,58 @@
+# BIC - Param Singh - The On-Chain Hero: Dynamic Rarity Action Figure Agent
+
+- **Proposer:** Param Singh
+- **Payment Address:** 0x5A43a975a632a77828BF9659F1f09Aa53C99C4B9
+- **Contact Email:** paramsingh1810@gmail.com
+- **Project Slug:** [PR Link]
+
+## Abstract
+
+The "On-Chain Hero" is a sophisticated AI Agent that transforms user selfies into high-fidelity, 3D-modeled digital action figures housed in branded, interactive collector boxes. Leveraging my experience in optimizing Deep Learning models and real-time inference, this agent performs two distinct tasks:
+
+1.  **Visual Feature Analysis:** Using a lightweight classification model (like MobileNet v2) to detect user attributes (e.g., "Developer," "Creator," "Explorer").
+2.  **Generative Transformation:** Applying Stable Diffusion 1.5 with custom-trained LoRAs and ControlNet to render the user as a high-quality vinyl figure within a branded box template featuring the Cere Network logo.
+
+The agent is "viral-ready" because it generates a unique "Rarity Score" for every user, encouraging social sharing and competition during partner campaigns.
+
+## Team 🧑‍🤝‍🧑
+
+- **Team Leader:** Param Singh
+- **Contact Email:** paramsingh1810@gmail.com
+- **GitHub Profile:** [ParamSingh24](https://github.com/ParamSingh24)
+
+### Team's Experience
+
+-   **SciPy Contributor:** Successfully rectified mathematical inconsistencies in the `sparse.linalg.eigsh` documentation (**PR #18214**), proving a deep understanding of low-level mathematical operator definitions.
+-   **Hackathon Champion:** 1st Place Winner of Hackathon 2025 (BBD University) and Winner of "AI and Robotics for Disaster Management" by WUST.
+-   **MLE & Full Stack Expert:** Experienced in architecting end-to-end ML pipelines for IoT and enterprise workflows, ensuring seamless frontend-backend integration.
+-   **Academic Excellence:** B.Tech in CSE (AI) with a 9.33 CGPA, consistently on the Dean’s List.
+
+## Technical Specification
+
+-   **Agent Logic:** Built using Python and PyTorch, the agent will ingest a photo, run a feature-detection pass, and dynamically construct a prompt (e.g., "A legendary developer action figure in a Cere-branded neon box").
+-   **Diffusion Pipeline:** Uses Stable Diffusion 1.5 + styled LoRA (for plastic texture) + ControlNet Canny/Depth (to preserve facial structure and box geometry).
+-   **Data Sovereignty:** All inference and model assets will be managed via Cere’s Decentralized Data Clusters (DDC), ensuring user data privacy—a key differentiator for this project.
+
+## Development Roadmap 🔩
+
+- **Total Estimated Duration:** 1 Month
+- **Full-Time Equivalent (FTE):** 1
+
+### Milestone 1 — Logic & Model Preparation
+- **Duration:** 2 Weeks
+- **Deliverables:**
+    -   **Feature Analysis Script:** A Python module using OpenCV/PyTorch to identify user "traits" for rarity assignment.
+    -   **LoRA Training:** Fine-tuning a style model for the "Collectible Toy" aesthetic.
+    -   **Branded Templates:** Designing the 3D-style box featuring the Cere logo and partner slots.
+
+### Milestone 2 — Cere Stack Integration & Deployment
+- **Duration:** 2 Weeks
+- **Deliverables:**
+    -   **DDC Content Storage:** Uploading model weights to Cere DDC and configuring the Model Registry.
+    -   **Telegram/Discord Integration:** Deploying the agent as a programmable unit that responds to the `/fun` command in whitelisted partner channels.
+    -   **Quality QA:** Ensuring results meet "ChatGPT-level quality" and consistent branding.
+
+## Future Plans
+
+-   **Dynamic Metadata:** Long-term, I plan to integrate the rarity scores as on-chain metadata for users to mint their action figures as NFTs.
+-   **Monetization:** Utilizing the $100 "per-win" bonus structure to fund continuous updates for new partner-specific box designs.