Creator prompt
The idea behind this presentation
Create a professional, visually clean academic B.Tech major-project presentation (12–14 slides) titled "A Blockchain-Verified Dynamic Controller for Multi-LLM Orchestration on Distributed Compute Clusters."
Context: This is submitted by Team Wolfs — Saksham Godiyal (2319484) and Anubhav Semwal (2318445), B.Tech CSE, Graphic Era Hill University, Dehradun, under the guidance of Ms. Manika Manwal (Project Team ID: CSE27-266).
Slide-by-slide structure:
Title Slide — Project title, team name (Team Wolfs), members, roll numbers, guide name, department, university.
Abstract — One-slide summary: multi-LLM orchestration problem, the existing MPI cluster foundation, and the two new layers (Dynamic Load Controller + Blockchain Verification).
Introduction & Background — Why multi-model AI coordination matters; the coordinator-worker analogy to MPI scatter/gather.
Motivation / Problem Statement — Two gaps: (a) no verifiable audit trail for routing/output decisions, (b) academic clusters stop at benchmarking without real deployment use cases.
Existing Foundation — The MPI Cluster — 4-node Raspberry Pi 5 + Jetson Nano cluster, coordinator/worker topology, MPI_Scatter/MPI_Gather, validated results: 6.13x speedup, 1600x less inter-node data transfer. Include the cluster topology diagram.
Literature Review Highlights — Briefly cite: Switch Transformer / Mixtral (MoE routing), RouteLLM (external routing), Model Cards / HASC (AI transparency), Hyperledger Fabric & Lightweight IoT Blockchain (edge verification) — and the gap each leaves.
Proposed System Architecture — Three layers: Model Council Layer, Dynamic Load Controller, Blockchain Verification Layer. Include the architecture flow diagram (Incoming Query → Controller → Model Council → Aggregation/Arbitration → Blockchain Layer → Verified Response).
Request Life Cycle — 5 steps: Ingestion → Routing → Parallel Inference → Aggregation/Arbitration → Verification & Response.
Blockchain Verification Design — Hash-chained audit log vs. permissioned ledger (Hyperledger Fabric) option; what each ledger entry stores.
Methodology / Implementation Stages — Stage 1–7 (cluster readiness → model deployment → controller → blockchain layer → integration → benchmarking → documentation).
Technology Stack — Hardware (Raspberry Pi 5 x3-4, Jetson Nano), Software (MPI/OpenMPI, OpenMP, Python, llama.cpp/Ollama, Flask/FastAPI dashboard, hash-chain/Hyperledger).
Roadmap & Milestones — 6-phase, 12-month timeline table.
Expected Outcomes — Working demo + dashboard, research paper, patent application, benchmarking report.
Conclusion — Reposition of validated MPI cluster into a trustworthy, auditable, edge-deployable multi-model AI system.
Design guidance: Use a clean academic/tech theme (dark navy + orange/teal accents to match the GEHU branding), consistent iconography for hardware/blockchain/AI, minimal text per slide (bullet fragments, not full sentences), and include the two provided diagrams (cluster topology, system architecture flow) as visuals on slides 5 and 7. Keep a technical, formal tone suitable for a faculty mentor review.