Bringing clarity to public services for everyone.
CivicLens is a chat-based platform that helps people find answers about government schemes, public services, and civic processes. Instead of digging through government websites, you can just ask questions in plain English and get clear answers.
Try it now: https://tryciviclens.vercel.app
Chat interface with streaming responses Ask questions about government schemes, eligibility, or how to apply for benefits. Responses stream in real-time using OpenAI's API. Works for multiple countries including India, USA, UK, Canada, and Australia.
File uploads Upload PDFs or documents related to government schemes, and the system extracts relevant information to help answer your questions.
Query insights Automatically extracts entities (schemes, offices, documents) from your questions and suggests relevant actions. Also classifies queries to better understand what you're looking for.
Kestra workflows Automated workflows that can crawl government portals, extract eligibility rules from PDFs, and generate autocomplete suggestions. These run via webhooks that integrate with the main app.
Oumi fine-tuning Custom dataset with 12 civic Q&A examples covering multiple countries. Can be used to fine-tune a model for better civic-specific responses, though it's optional and the app works fine with standard OpenAI models.
- Next.js 16 - React framework
- MongoDB - Database (via Prisma)
- Clerk - Authentication
- OpenAI - AI chat responses (GPT-3.5-turbo)
- Kestra - Workflow automation
- Oumi - Model fine-tuning (optional)
- TailwindCSS + shadcn/ui - UI components
- Vercel - Deployment
The main AI engine. Handles streaming chat responses, query classification, entity extraction, and document generation (RTI/FOIA requests, complaint letters, eligibility summaries). Code is in lib/ai/ and app/api/ai/.
Three workflows for automating government data tasks:
- Weekly scheme updates (crawls government portals)
- Eligibility rule extractor (parses PDFs)
- Autocomplete suggestions (matches queries to schemes)
Workflows are in kestra/workflows/ and integrate via webhook endpoints in app/api/kestra/webhook/.
Fine-tuning setup for a custom civic model. Dataset has 12 examples covering India, USA, UK, Canada, and Australia. Config is in oumi/config.yaml. The app can use a fine-tuned model if available, but falls back to standard OpenAI if not.
Used during development for code generation, refactoring, and multi-step feature work. .clinerules file contains project-specific guidelines. There's a sample workflow script at scripts/cline-workflow.sh.
Deployed on Vercel with edge functions for low-latency responses. GitHub integration handles automatic deployments.
- Node.js 18+
- MongoDB (local or Atlas)
- OpenAI API key
- Clerk account (for auth)
- Install dependencies:
npm install- Set up environment variables:
cp .env.example .env
# Edit .env with your keysRequired variables:
OPENAI_API_KEY- OpenAI API keyDATABASE_URL- MongoDB connection stringCLERK_SECRET_KEY- Clerk secret keyNEXT_PUBLIC_CLERK_PUBLISHABLE_KEY- Clerk public keyKESTRA_URL- Kestra instance URL (optional)KESTRA_AUTH- Base64 encoded Kestra credentials (optional)
- Set up the database:
npx prisma generate
npx prisma db push- Run the dev server:
npm run dev- (Optional) Set up Kestra:
- Install Kestra: https://kestra.io/docs/getting-started
- Deploy workflows from
kestra/workflows/ - Configure CORS using
kestra/application.yml
- (Optional) Fine-tune with Oumi:
pip install oumi[gpu]
oumi train -c oumi/config.yamlapp/- Next.js app router pages and API routescomponents/- React components (UI, chat, file upload)lib/- Core logic (AI services, database, file extraction)hooks/- React hooks for chat operationsprisma/- Database schemakestra/workflows/- Kestra workflow definitionsoumi/- Fine-tuning config and dataset
The app uses streaming responses for chat, which provides a better user experience. File uploads are stored in MongoDB (base64 encoded for small files). The Kestra workflows are separate and can be deployed independently — they communicate with the main app via webhooks.
MIT