All work

■ 02 · Agentic AI / RAG / PostgreSQL

Municipal Asset Intelligence Copilot

Agentic AI · RAG · Enterprise AI

May 2026 — Jul 2026

Overview

An Agentic AI-powered intelligence copilot designed to perform accurate multi-step reasoning across millions of municipal asset records. Built during the Agentic AI Engineer internship at EPIC Technology Research.

Problem

Municipal asset data lives across millions of records in heterogeneous formats. Keyword search cannot answer multi-step operational questions that require intent understanding, retrieval over structured + vector data, and reasoned synthesis.

Approach

  1. 01Intent-aware query routing to direct each question to the right retrieval path
  2. 02Hybrid RAG over PostgreSQL + pgvector combining structured filters with vector search
  3. 03On-prem LLM orchestration for multi-step reasoning with retrieved context

Architecture

  1. USER QUERY
  2. INTENT DETECTION
  3. QUERY ROUTER
  4. RETRIEVAL LAYERPostgreSQL · pgvector · Hybrid Search
  5. LLM ORCHESTRATIONOn-prem
  6. MULTI-STEP REASONING
  7. ANSWER
Conceptual flow — components limited to what the resume explicitly supports.

Engineering

Intent-aware query routingHybrid RAGPostgreSQLpgvectorOn-prem LLM orchestration
  • Intent-aware routing
  • Hybrid retrieval
  • Multi-step reasoning
  • Large-scale asset Q&A

Results

MillionsAsset records reasoned over

View on GitHub