(Note: Public demo TBD)

🔧 Keywords

Report Automation, Performance Summaries, Database Querying, EdTech, SQL Agents, RAG, Multi-Agent Systems, LangChain, Milvus, NLP, Modular Pipelines

🧩 Problem

The client, an EdTech platform, needed a chatbot to help program managers quickly analyze student performance. While the platform’s data (grades, assignments, materials, feedback) was already structured, relevant insights were scattered across multiple tables and sources. Manually piecing together this information for reporting was time-consuming, inconsistent, and hard to scale.

They needed a system that could generate detailed, personalized reports quickly and at scale.

⚙️ What I Did

• Designed and implemented a modular multi-agent architecture with a Router Agent, RAG Agent, and SQL Agent.
• Built the RAG Agent to generate personalized answers using embedded student-level data stored in Milvus, powered by OpenAI embeddings.
• Created a custom ETL pipeline to extract relevant content from a PostgreSQL database, including grades, submission history, and instructor feedback.
• Structured and embedded that data with rich metadata (e.g. user ID, activity type, grading notes) to support context-aware retrieval.
• Contributed to the Router Agent logic that classifies each question as SQL or vector-based, ensuring correct agent routing.
• Used LangChain to orchestrate the pipeline and support dynamic tool usage per query type.

📊 Outcome

• Delivered a system that automates both SQL-based queries and semantic search for performance summaries, improving relevance and accessibility.
• Reduced manual effort for support teams and enabled faster, more accurate student reports — with just 2 input fields (student name + question).
• Helped the client explore a scalable AI support layer for future LMS-wide rollout.