HUY.
AI × BLOCKCHAIN2024Full-stack Developer

D-CERT

A digital diploma platform combining blockchain verification with an AI-powered knowledge retrieval system.

D-CERT addresses the challenges of academic credential fraud and complex institutional document lookup by uniting immutable blockchain verification with an advanced retrieval-augmented generation (RAG) assistant.

The Problem

Verifying academic degrees is traditionally slow and vulnerable to tampering. Simultaneously, navigating institutional regulations, graduation guidelines, and academic records involves labor-intensive manual searching.

The Solution

Constructed a dual-engine platform: smart contracts on Ethereum Sepolia store document hashes for instant, tamper-proof diploma verification; while a Python FastAPI + FAISS vector engine indexes institutional PDFs and queries a local Qwen 2.5 14B LLM for accurate context-grounded answers.

Hybrid microservices architecture: React client interfaces with a Node.js/Express application core for business workflows, an Alchemy-connected smart contract for on-chain verification, and a FastAPI RAG microservice for semantic vector search.

Execution Flow & Pipeline:
01Document Hash Pipeline: Certificate PDF → SHA-256 Hash → bytes32 → Smart Contract on Ethereum Sepolia
02RAG Pipeline: Institutional PDF → PyMuPDF → Semantic Chunking → BKAI Bi-Encoder Embeddings → FAISS Vector Store
03Inference Flow: User Question → Dense Vector Query → FAISS Retrieval → Prompt Augmentation → Qwen 2.5 14B (Ollama) → Verified Answer + Sources

Tamper-Proof Verification

On-chain verification verifying SHA-256 certificate hashes directly against Ethereum smart contracts.

Semantic RAG Search

Specialized Vietnamese embedding model (BKAI Bi-Encoder) coupled with FAISS for semantic document chunk lookup.

Context-Grounded LLM Responses

Qwen 2.5 14B provides hallucination-resistant answers with precise citations to source document pages.

Administrative Portal

Role-based certificate issuance, student record management, and batch document processing.

CHALLENGE:Accurate Vietnamese Semantic Retrieval

Approach: Benchmarked and integrated the BKAI Vietnamese Bi-Encoder to generate high-quality dense vector representations for Vietnamese academic terminology.

CHALLENGE:Gas Optimization & Hash Integrity

Approach: Stored document SHA-256 hashes as compact bytes32 types in Solidity rather than raw string data, reducing on-chain gas costs significantly.

Deployment

FastAPI service running with local Ollama inference; Node.js core backend; smart contracts deployed to Ethereum Sepolia via Alchemy RPC.

Result

Demonstrated instantaneous diploma authenticity verification and highly accurate Vietnamese academic document querying with zero hallucination.

Technologies Used

ReactTailwind CSSNode.jsExpressMongoDBFastAPIFAISSBKAI Vietnamese Bi-EncoderQwen 2.5 14BOllamaSolidityEthereum SepoliaEthers.jsAlchemy

Personal Contribution

  • Architected the full system integration between Web2, AI service, and Web3
  • Developed the FastAPI AI microservice with FAISS and Ollama LLM integration
  • Authored and deployed the Solidity certificate registry contract to Sepolia testnet
  • Implemented the web frontend and credential lookup portal