fixora
AI-Powered Civic Infrastructure Reporting with Multi-Agent Verification & Vector RAG
Project Overview
An open civic data and public infrastructure reporting platform powered by AI. Frictionless without compulsory login, verified by Multi-Agent LLMs and Qdrant Vector DB.
- —Frictionless zero-login reporting maximizing civic participation while safeguarding whistleblower identity.
- —Multi-Agent LLM verification pipeline (Advocate, Skeptic, Manager) filtering spam and verifying legitimacy.
- —Semantic matching of regional municipal budgets using high-dimensional Qdrant Vector Database.
- —Golang Modular Monolith backend (Fiber + GORM) achieving sub-50ms REST endpoint latencies.
The Challenge
Conventional municipal channels suffer from cumbersome registration hurdles, weeks of manual verification lag, and total opacity in infrastructure budget allocation.
- —Low citizen participation caused by mandatory account creation and privacy exposure concerns.
- —Flood of duplicate tickets, photo spam, and irrelevant submissions overwhelming manual reviewers.
- —Municipal budget documents (APBD) trapped in static PDFs that defy manual cross-referencing with field incidents.
- —Backend needed high-throughput ingestion of high-resolution incident photos during severe weather surges.
The Solution
Engineered an anonymous encrypted intake workflow backed by a multi-stage AI verification pipeline and semantic budget cross-referencing.
- —Zero-Login reporting flow utilizing ephemeral cryptographic session tokens to ensure complete whistleblower anonymity.
- —Computer Vision preprocessing classifying pavement damage severity and municipal utility types.
- —Multi-Agent LLM consensus where Advocate synthesizes evidence, Skeptic hunts anomalies, and Manager decides final verification.
- —Qdrant Vector Database indexing municipal fiscal datasets to instantly correlate repairs with allocated public funding.
Key Features
Functional capabilities built to solve core user and operational friction.
Zero-Login Anonymous Reporting
Citizens can capture and dispatch infrastructure incidents in under 30 seconds without creating accounts or risking privacy leaks.
Multi-Agent AI Verification
Three discrete AI agents (Advocate, Skeptic, Manager) independently evaluate image features, GPS metadata, and textual context.
Qdrant Semantic Budget RAG
Matches damaged infrastructure coordinates with West Java public works budget line-items via high-speed semantic vector search.
Interactive Geospatial Heatmap
Live visualization of public facility status, repair progress, and district severity clusters open to public scrutiny.
Autonomous News & Data Crawler
Autonomously ingests incident signals from regional news feeds and open data portals to enrich community-submitted reports.
Modular Monolith Architecture
Reporting, authentication, AI pipelines, and analytics domains isolated cleanly inside a unified high-performance Go binary.
System Architecture
Architectural patterns, backend performance paradigms, and key design decisions.
Engineering Process
Development lifecycle from modeling to production deployment.
System Modeling & AI Pipeline Design
Modeled zero-login data privacy, vector embeddings for municipal budgets, and agent consensus rules.
- —Reporter privacy threat modeling
- —Budget vector indexing in Qdrant
- —Agent persona & validation guardrails
Golang Backend & Clean Architecture Build
Built high-throughput REST endpoints in Go Fiber with modular domain isolation and CV inference.
- —Clean Architecture repository implementations
- —Resilient image ingest pipeline with auto-compression
- —Asynchronous workers for AI consensus inference
Containerization & Production Deployment
Containerized deployment orchestrated via Docker with stress-testing under simulated traffic bursts.
- —Multi-stage Docker container optimization
- —Qdrant vector query latency tuning
- —Public beta launch & community verification