CivicBridge_AI

Connecting citizens to verified government opportunities-Faster.

Comment

Inspiration

India runs hundreds of scholarships, welfare schemes, internships, and grants — yet for most citizens the path looks like a maze. Information is scattered across MyScheme, the National Scholarship Portal, UMANG, DigiLocker, state sites, and PDF notifications that read like legal code. Students miss deadlines. Families upload the wrong documents. Farmers and first-time founders fall for middlemen who “help†for a fee.

We were inspired by a simple frustration: eligibility should not be a privilege of people who already know where to look. CivicBridge AI was born from the idea that a citizen should describe who they are once — education, income band, state, skills — and immediately see what they can apply for, what documents they need, and what to do next, without leaving one trusted surface.

Our north star was not “another chatbot.†It was a bridge: verified government opportunities on one side, clear AI-guided action on the other.

What it does

CivicBridge AI is an opportunity-intelligence platform for Indian citizens. It helps students, job seekers, founders, farmers, and families discover scholarships, internships, welfare schemes, and grants — then guides them through eligibility, documents, roadmaps, and deadlines.

Capability What users get
Verified schemes $40+$ real central schemes with official portal links (NSP, PM-KISAN, PM-JAY, PMAY, SIH, Startup India, PMKVY, and more)
Smart eligibility Profile-based match scores with plain-language reasons
Document AI Dense government notifications explained in clear steps
Application roadmaps Step-by-step plans grounded in official scheme processes
Deadline calendar Real dates across scholarships, grants, and welfare schemes
Secure vault Private document storage (Firebase Storage when signed in)
Career advisor Skills, certifications, and projects that strengthen eligibility
AI assistant Chat about schemes, documents, and next steps via Gemini

Eligibility and AI guidance are advisory. Users always confirm details and submit on the official government portal.

Live app: https://civicbridge-ai.vercel.app

How we built it

We split the product into three layers that talk over a clean API boundary.

Frontend

  • Next.js 16 (App Router), React 19, TypeScript
  • Tailwind CSS + shadcn/ui + Framer Motion for a fast, accessible dashboard
  • Client modules for eligibility scoring, scheme search, vault UX, and optimistic AI fallbacks so the UI never feels blocked

Backend (AI)

  • FastAPI service exposing /health and /api/ai/*
  • Google Gemini (google-genai) for career advice, document explanation, roadmaps, and chat
  • Deterministic fallback responses when the model is slow, missing, or times out — so demos and production stay reliable

Auth & data

  • Firebase Authentication for signup / login
  • Firestore for vault metadata and user profile data
  • Firebase Storage for document uploads (with client-side image compression and progress)

Deployment

  • Single Vercel project hosting Next.js + FastAPI as services
  • Same-origin API in production (/api, /health) to avoid CORS friction
  • Secrets loaded as environment variables — never committed to GitHub

At a high level, a user’s match relevance can be thought of as a weighted blend of profile fit and deadline urgency:

$$ S(u, o) = \alpha \cdot E(u, o) + \beta \cdot D(o) + \gamma \cdot T(u, o) $$

where $E(u,o)$ is eligibility fit for user $u$ and opportunity $o$, $D(o)$ encodes deadline proximity, $T(u,o)$ captures tag/interest overlap, and $\alpha + \beta + \gamma = 1$. The product surfaces $S(u,o)$ as an interpretable match score with human-readable reasons — not a black box ranking.

Challenges we ran into

  1. Fragmented, messy real-world scheme data
    Official portals differ in structure, update cadence, and wording. We curated a verified scheme set with official links and treated AI output as guidance, never as a substitute for the government source of truth.

  2. AI latency vs. demo-quality UX
    Cold Gemini calls can take several seconds. Career Advisor and Document Vault originally felt “stuck.†We fixed this with instant local/cached advice, background refinement, upload progress, image compression, and strict request timeouts.

  3. Monorepo deployment
    Hosting Next.js and FastAPI together on Vercel Services required careful vercel.json setup, same-origin API wiring, and health checks so the frontend and AI backend stay aligned in production.

  4. Auth on the live domain
    Firebase rejects unknown hosts. Production login needs civicbridge-ai.vercel.app added to Firebase Authorized domains — a classic “works locally, breaks live†footgun.

  5. Trust & safety
    Citizens can be harmed by invented URLs or overconfident eligibility claims. We constrained prompts toward real portals (MyScheme, NSP, DigiLocker, etc.), added fallbacks, and put a clear advisory disclaimer in the product.

Accomplishments that we're proud of

  • A working end-to-end product, not just a slide deck — live at civicbridge-ai.vercel.app
  • $40+$ verified schemes with official portals, not scraped junk links
  • Eligibility + roadmaps + document AI + vault + career advisor + chat in one coherent citizen journey
  • Graceful degradation: if Gemini is slow or unavailable, the app still helps
  • Clean open-source hygiene: MIT license, security docs, env templates without secrets on GitHub
  • Shipping performance fixes under time pressure so vault uploads and career advice feel instant

What we learned

  • Civic tech is a trust problem first, a UI problem second. People will only follow guidance if every link and claim can be traced to an official source.
  • Optimistic UX beats waiting for the perfect model response. Showing a solid local answer in $t \approx 0$, then refining asynchronously, is often better than a spinner for $5$–$10$ seconds.
  • Fallbacks are a feature. Deterministic templates for career advice, roadmaps, and chat keep the product usable when APIs flake.
  • Deployment is part of the product. Same-origin APIs, health checks, and Firebase authorized domains matter as much as feature code.
  • Building for India means designing for language density, document friction, and deadline anxiety — not generic “productivity app†patterns.

What's next for CivicBridge_AI

  • Deeper scheme coverage — more state schemes, better refresh pipelines from official sources
  • Multilingual support — Hindi and major regional languages for explanations and chat
  • Smarter eligibility — structured rules engines alongside Gemini for auditable match reasons
  • Proactive alerts — push / email reminders before deadlines and missing-document nudges
  • DigiLocker-aware flows — guide users toward official document fetches where possible
  • NGO & campus partnerships — help counselors run CivicBridge as a shared discovery desk
  • Accessibility & low-bandwidth modes — lighter pages for users on slower networks

CivicBridge’s long-term goal is simple: every eligible citizen should find the right opportunity in minutes, not months — and never need a middleman to do it.

Built With

Share this project:

Updates