AI-Powered DPR Intelligence & Evaluation Platform
Multi-agent evaluation engine for infrastructure DPRs with Next.js frontend, FastAPI AI backend, geospatial risk analysis, RAG search, and automated document intelligence.
VOIDFRAME2 is an intelligent platform for Detailed Project Report (DPR) analysis and decision support, designed for infrastructure, public sector, and project evaluation workflows.
The system combines:
- Next.js Frontend for portal, dashboards and interactive workflows
- FastAPI + Python AI Backend for autonomous DPR analysis
- Multi-Agent LLM Evaluation (Engineering, Finance, Risk, Policy, Reviewer)
- RAG + FAISS Semantic Search over uploaded DPRs
- GIS & Risk Intelligence for terrain, flood, landslide and climate analysis
- BoQ + Compliance Analysis for financial and policy checks
- Annotated PDF Output highlighting identified issues
Upload a DPR → AI agents analyze it → risks/compliance/cost issues are surfaced → users receive:
- AI evaluation report
- Feasibility scoring
- Compliance gaps
- Cost anomalies
- Risk simulations
- GIS insights
- Annotated DPR with flagged concerns
- Interactive question-answering over the DPR
Five specialized AI agents collaborate:
Analyzes:
- Technical feasibility
- Design adequacy
- Slope, drainage, structural concerns
- Engineering risks
- Mitigation recommendations
Evaluates:
- BoQ and cost structures
- Rate anomalies
- Budget concentration risks
- Cost benchmarking
- Financial feasibility
Quantifies:
- Project execution risk
- Terrain/weather uncertainty
- Monte Carlo simulations
- Cost/timeline overrun probability
Checks:
- Guideline compliance
- Institutional readiness
- Regulatory alignment
- Implementation preparedness
Synthesizes all agent outputs into:
- Final feasibility recommendation
- Consolidated scoring
- Approval / conditional / reject recommendation
- Executive summary
Ask questions directly against uploaded DPRs:
- DPR-grounded answers
- FAISS vector retrieval
- LLM synthesis
- Optional web validation
- Source-backed responses
Example:
“What contingency is budgeted for landslide mitigation?”
Integrated geospatial assessment:
- Elevation analysis
- Slope assessment
- Landslide risk
- Flood vulnerability
- Infrastructure proximity mapping
- Climate profile signals
Monte Carlo modeling:
- P10 / P50 / P90 scenarios
- Cost overrun probabilities
- Schedule risk estimation
- Contingency stress testing
Outputs issue-marked DPR PDFs with:
- Severity highlighting
- Issue annotations
- Review comments
- Summary pages
- Next.js
- React
- TypeScript
- TailwindCSS
- PostCSS
- FastAPI
- Python
- LangChain
- FAISS
- Sentence Transformers
- Groq / LLM inference
- Multi-Agent Reasoning
- Retrieval Augmented Generation (RAG)
- Monte Carlo Simulation
- Geospatial Risk Analysis
- Cost Benchmarking
- Compliance Scoring
VOIDFRAME2/
│
├── src/ # Next.js frontend
│ ├── app/
│ ├── components/
│ ├── hooks/
│ ├── services/
│ ├── lib/
│ └── types/
│
├── public/ # static assets
│
├── backend/ # FastAPI + AI engine
│ ├── api/
│ ├── ai/
│ ├── evaluations/
│ └── tests/
│
├── docs/
│ ├── Architecture/│
│ └── Backend/
│
├── package.json
├── next.config.ts
├── tsconfig.json
└── vercel.json
User uploads DPR PDF
↓
Frontend submits to FastAPI
↓
PDF Extraction
↓
FAISS Vector Store Build
↓
5 AI Agents Evaluate
↓
Additional Modules:
- GIS Engine
- Risk Simulation
- BoQ Parser
- Compliance Checks
- PDF Annotation
↓
Unified Evaluation Object
↓
Results returned to Frontend
The frontend uses a unified semantic UI system featuring:
- Semantic color system
- Typography scale
- Spacing system
- Reusable card system
- Button variants
- Glass/surface components
- Shared layout patterns
Core reusable classes include:
page-wrapper
page-content
card-default
surface-glass
surface-glass-md
btn-primary
btn-secondary
btn-ghost
input-base
heading-xl → heading-sm
body-lg → body-xs
See documentation in /docs/style-guides.
POST /upload_dprRuns full evaluation pipeline.
POST /askRAG-powered question answering.
POST /evaluate_dprGET /healthgit clone <repo-url>
cd VOIDFRAME2npm install
npm run devRuns:
http://localhost:3000
cd backend/Pyhton-Evaluation-Backend
python -m venv venv
source venv/bin/activate
pip install -r requirements.txtRun FastAPI:
uvicorn app:app --reload --port 8000Backend:
http://localhost:8000
Create .env
GROQ_API_KEY=your_key
SERPER_API_KEY=your_key
AGENT_MODEL=llama-3.1-8b-instantFrontend variables (if needed):
NEXT_PUBLIC_API_URL=http://localhost:8000User uploads PDF.
Pipeline runs:
- AI agent review
- Risk quantification
- GIS checks
- Compliance scoring
- Annotation
Returns:
{
"evaluation": "AI report...",
"issues": [],
"gis_analysis": {},
"risk_summary": {},
"highlighted_pdf": "annotated.pdf"
}See:
docs/architecture/
Contains:
- Design System
- Visual Style Guide
- UI Uniformity Implementation
- Changes Summary
See:
docs/backend/
Includes:
- Backend Summary
- Function Breakdown
- Complete Code Flow
- Quick Reference
- Visual Architecture
- Documentation Index
- Multi-agent evaluation
- RAG search
- Vector search
- Compliance engine
- BoQ parsing
- Monte Carlo simulation
- GIS engine
- PDF annotation
- Portal dashboards
- Upload workflows
- Chat assistant
- Result visualizations
- Unified UI system
This codebase follows:
- Modular architecture
- Service-oriented backend layers
- Shared design system
- Reusable component patterns
- Semantic styling
- Documentation-first development
Planned/possible extensions:
- Autonomous DPR review agents
- Multi-DPR comparative analysis
- Workflow approvals
- Government scheme benchmark engine
- Agentic remediation suggestions
- Procurement intelligence
- Predictive project failure scoring
Common commands:
npm run dev
npm run build
npm run lintuvicorn app:app --reload
pytestSuggested workflow:
git checkout -b feature/my-featureFollow:
- unified design system
- backend layered architecture
- linting rules
- documentation updates with changes
Current focus:
- AI DPR Evaluation Platform
- Frontend + FastAPI integration
- Multi-agent infrastructure intelligence
VOIDFRAME2 aims to turn DPR review from a manual document-heavy exercise into an AI-assisted intelligence workflow for faster, better infrastructure decisions.
Built with:
- Next.js
- FastAPI
- LangChain
- FAISS
- Multi-Agent AI