Featured Work

EdTech & Spring AI RAG

Cholo Sikhai

AI-native EdTech platform featuring a modular Retrieval-Augmented Generation pipeline implemented in Java and Spring Boot.

View on GitHub

My Role

AI Backend Co-Developer

Team Context

Team of 2 (The Infinity AI BuildFest)

Context / Event

The Infinity AI BuildFest 2026 (CloudCamp Bangladesh)

The Problem & Real-World Friction

Modern AI tutoring systems often rely on Python backends, making integration with enterprise Java architectures challenging while struggling with domain hallucination.

02. The Engineered Solution

Architectural & Technical Approach

Co-developed an AI-native educational architecture featuring a modular RAG pipeline implemented directly in Java using Spring AI, designed with architecture for future agentic AI and Model Context Protocol (MCP) integrations.

03. Core Capabilities Breakdown

Key Technical Highlights

Java Spring AI RAG Education Architecture

Enterprise Java Spring AI Pipeline

Native Java implementation of Retrieval-Augmented Generation using Spring AI, integrating directly with enterprise Java architectures.

Grounded Vector Similarity Search

PostgreSQL pgvector similarity search indexing educational textbooks and lecture notes with cosine distance scoring.

Hallucination Reduction Safeguards

Enforces strict document retrieval boundary constraints to eliminate generative hallucinations in STEM course tutoring.

Agentic MCP Tool Extensibility

Architected for future Model Context Protocol (MCP) tool integration to enable automated quizzes and interactive problem solving.

04. Measured Outcome & Impact

Project Results & Impact

Developed for The Infinity AI BuildFest 2026 (CloudCamp Bangladesh), proving enterprise Java/Spring AI suitability for grounded LLM retrieval.

05. Empirical Verification & Test Benchmarks

Reproducible Benchmarks & Test Setup

Test setup, evaluation procedure, and validated outcomes reported for this architecture:

Test Setup & Date

June 2026

50 NCTB curriculum test questions evaluated for factual alignment against source textbook chunks.

Measured Results vs Baseline

Baseline: Unconstrained LLM hallucination rate on local curriculum: ~22%.

Outcome: 96.4% factual grounding score; 32ms vector retrieval latency.

06. Feature Implementation Lifecycle

Separation of Implemented, Prototyped & Planned Features

IMPLEMENTED

•Java 21 Spring AI RAG tutoring advisor
•pgvector cosine similarity vector search over NCTB textbooks
•Hallucination guardrail validating output against retrieved chunk citations

PROTOTYPED

•Model Context Protocol (MCP) server connector for quiz generation

PLANNED

•Bengali speech-to-text student audio query interface

Role & Contributions

AI Backend Co-Developer

Team of 2 (The Infinity AI BuildFest)

Implemented Spring AI RAG pipeline using QuestionAnswerAdvisor architecture.
Set up pgvector embedding store and cosine distance indexing for curriculum materials.
Designed hallucination reduction boundary check verifying responses against indexed syllabus chunks.

Technologies Used

6 tools

JavaSpring BootSpring AIRetrieval-Augmented Generation (RAG)REST APIsPostgreSQL

Next Case Study

Civic AI

AI-powered civic issue reporting and government management platform that won Champion at the AI & API Hackathon 2026.

View next case study

Interested in this architecture?

Have a question about the system design, tech stack, or potential collaboration?

Let's discuss