AI-Powered RFI Resolution
Deploys an 8-agent autonomous system that classifies RFIs using NLP, searches documents via RAG, detects BIM clashes, routes to optimal experts, analyzes schedule/cost impact, drafts responses with citations, validates quality, and updates knowledge bases—reducing resolution time from days to minutes..
Problem Statement
The challenge addressed
Solution Architecture
AI orchestration approach
RFI Submission - Input form with sample RFIs, discipline selection, and 8-agent pipeline overview showing NLP, RAG, and BIM capabilities.
Multi-Agent Processing - Live orchestration view showing agent reasoning, classification results, and real-time activity feed with cost tracking.
Executive Summary - Analysis results with business impact, resolution timeline, AI-detected cost savings, and prioritized action items.
ESG & Sustainability View - Carbon footprint analysis, LEED compliance tracking toward Gold certification, and gap analysis recommendations.
AI Agents
Specialized autonomous agents working in coordination
NLP Classifier Agent
RFIs arrive with varying formats and terminology. Manual classification and prioritization is inconsistent and time-consuming, leading to misrouting and delays.
Core Logic
Uses TF-IDF combined with BERT embeddings (distilbert-base) for semantic understanding. Classifies RFIs by discipline, urgency, and complexity. Extracts named entities (persons, locations, building systems), performs sentiment analysis, and assigns confidence scores. Outputs structured classification for downstream routing.
RAG Searcher Agent
Finding relevant specifications, drawings, and historical RFI resolutions across thousands of project documents is a manual needle-in-haystack exercise.
Core Logic
Uses Pinecone vector database with text-embedding-3-small for semantic search. Performs cosine similarity matching against project specifications, drawings, and historical RFI database. Returns ranked results with relevance scores, excerpts, and page references. Identifies similar past RFIs with their resolutions for precedent-based answering.
BIM Clash Detector Agent
Many RFIs stem from coordination issues between disciplines (MEP vs. structural, architectural vs. fire protection). Manual clash detection in BIM models is slow and error-prone.
Core Logic
Uses Octree spatial indexing algorithm with IFC parser for BIM model analysis. Detects three clash types (hard, soft, clearance) across building systems. Identifies clash location, severity, and affected elements. Generates visualization data for 3D clash viewing. Provides resolution suggestions based on clash type and systems involved.
Expert Router Agent
RFIs often go to the wrong person or sit in queues. Matching RFI content to the right expert based on skills, availability, and workload is complex.
Core Logic
Uses collaborative filtering with matrix factorization for expert matching. Analyzes RFI requirements against expert skill profiles, current workload, and response time history. Recommends optimal expert assignment with estimated response time. Supports priority escalation rules and backup routing when primary experts are unavailable.
Impact Analyzer Agent
RFIs often have hidden schedule and cost implications. Without impact analysis, teams make decisions without understanding downstream consequences.
Core Logic
Uses CPM logic and Monte Carlo simulation for impact assessment. Calculates schedule impact in days based on activity dependencies, estimates cost impact from labor, materials, and delay costs, assesses resource implications, and computes risk scores. Recommends mitigation strategies to minimize RFI impact on project performance.
Response Drafter Agent
Drafting technically accurate RFI responses with proper citations and clear language requires significant expert time and writing skill.
Core Logic
Uses GPT-4 Turbo with RAG context for response generation. Synthesizes findings from all upstream agents into clear, technically accurate response drafts. Includes citations to source documents (specs, drawings, standards), confidence scoring, and recommended attachments. Produces ready-for-review responses that maintain consistent quality and format.
QA Validator Agent
RFI responses may contain errors, omit critical information, or fail to address the original question completely. Manual QA is inconsistent.
Core Logic
Uses rule engine combined with NLU for comprehensive validation. Performs compliance checks against response requirements, validates completeness against original RFI question, verifies citation accuracy, and cross-references against project standards. Assigns quality scores and flags issues requiring revision before approval.
Knowledge Curator Agent
Valuable knowledge from RFI resolutions is lost. Similar questions get asked repeatedly because there's no systematic learning from past responses.
Core Logic
Uses Neo4j knowledge graph with vector embeddings for knowledge management. Captures resolved RFIs with full context, identifies patterns across RFI types, updates knowledge base for future RAG retrieval, and builds relationships between project elements. Enables continuous improvement through accumulated project intelligence.
Worker Overview
Technical specifications, architecture, and interface preview
System Overview
Technical documentation
Tech Stack
6 technologies
Architecture Diagram
System flow visualization