Redressa

Agentic workflow that turns messy consumer complaint evidence into grounded, escalation-ready claim packages.

Hackathon: Protex Hack-2-Win 2026
Track: Track 2 – Agentic AI / AI Workflows

Stack

Layer	Tech
Frontend / App	Next.js 16 App Router + TypeScript + Tailwind v4
Backend	Server-side logic inside Next.js route handlers and server components
Database + Storage	Supabase (PostgreSQL + Storage)
AI Reasoning	Groq API (llama-3.3-70b-versatile)
OCR / Document Extraction	OCR.space
Deployment	Vercel

What It Does

1. Intake – The consumer describes their complaint and uploads supporting evidence (emails, screenshots, receipts).
2. Evidence Extraction – OCR and document parsing extract structured facts from uploaded files.
3. Policy Retrieval – The pipeline retrieves relevant company policies and consumer regulations (DGCA rules, E-Commerce Rules 2020) grounded to the complaint.
4. AI Analysis – Groq-powered reasoning evaluates the claim against retrieved policies, identifies violations, and scores claim strength.
5. Output Generation – The system produces a case summary, drafted grievance emails, escalation notes, and an evidence checklist.
6. Communication & Escalation – Built-in email drafting and escalation workflows with manual, assisted, and autopilot modes.

Current Baseline

• The live pipeline uses Groq as the primary reasoning provider.
• OCR.space is used for uploaded image OCR and best-effort PDF extraction.
• Support Breadth: The platform applies a Hybrid Grounding Model.
  • Merchant-grounded support: IndiGo and Flipkart have full policy coverage in the stable baseline.
  • Generic fallback: For unsupported merchants, the pipeline relies on ingested regulations and generic escalation guidance.
• The app provides informational guidance, not legal advice.

Retrieval Evolution Note

Currently, the pipeline isolates policy grounding through deterministic keyword matching and merchant-name normalization. Future Roadmap: In future versions, policy retrieval should migrate from keyword filtering to deep semantic embeddings stored in PostgreSQL using pgvector. A vector-based retriever would intelligently match conceptually identical grievances without requiring hardcoded keyword maps, dramatically improving retrieval accuracy on generalized merchant policies.

Getting Started

Prerequisites

• Node.js 18+
• npm
• A Supabase project
• A Groq API key
• An OCR.space API key

Local Setup

npm install
npm run dev

Open http://localhost:3000.

Environment Setup

Create .env.local from the example:

Copy-Item .env.example .env.local

Then fill in:

Variable	Where to find it
NEXT_PUBLIC_SUPABASE_URL	Supabase Dashboard > Project Settings > API > Project URL
NEXT_PUBLIC_SUPABASE_ANON_KEY	Supabase Dashboard > Project Settings > API > anon/public key
SUPABASE_SERVICE_ROLE_KEY	Supabase Dashboard > Project Settings > API > service_role key
GROQ_API_KEY	Groq Console
OCR_SPACE_API_KEY	OCR.space

Supabase Setup

1. Create a Supabase project.
2. Open the SQL Editor.
3. Run supabase/schema.sql.
4. Confirm the evidence storage bucket exists.
5. Confirm storage policies are present for the browser upload flow.
6. Visit /api/health locally and confirm Supabase is configured.

Project Structure

app/                    # Next.js pages, layouts, route handlers
  api/health/           # Health check endpoint
  api/pipeline/run/     # Pipeline execution endpoint
  api/case/delete/      # Secure claim deletion endpoint
  case/[id]/            # Case workspace (resizable panels)
  new/                  # Intake workspace
  profile/              # User profile management
lib/
  document/             # OCR.space adapter
  groq/                 # Groq reasoning adapter
  pipeline/             # Pipeline steps and orchestration
  supabase/             # Supabase clients and helpers
  types/                # Shared types
supabase/               # Database schema and migrations
data/
  policies/raw/         # Policy and regulation corpus

Key Features

• Resizable Workspace – Desktop case view uses react-resizable-panels for adjustable panel sizing; mobile auto-stacks.
• Communication Modes – Manual, assisted, and autopilot escalation workflows.
• User Profile – Editable name, email, and phone used as default variables in drafted complaint letters.
• Secure Claim Deletion – User-scoped deletion with ownership verification, confirmation step, cascading DB cleanup, and evidence storage pruning.
• Responsive Design – Premium dark editorial theme with mobile-optimized spacing and layout.

Deployment

Full guide: DEPLOYMENT.md

Quick version:

1. Push the repo to GitHub.
2. Import it into Vercel.
3. Add these environment variables before deploying:

Variable	Required
NEXT_PUBLIC_SUPABASE_URL	Yes
NEXT_PUBLIC_SUPABASE_ANON_KEY	Yes
SUPABASE_SERVICE_ROLE_KEY	Yes
GROQ_API_KEY	Yes
OCR_SPACE_API_KEY	Yes

4. Deploy.
5. Verify with /api/health.

Limitations

• OCR.space free-tier constraints still apply.
• Uploaded image OCR is verified working.
• PDF extraction is best-effort and may be constrained by file size and free-tier limits.
• The product provides informational guidance, not legal advice.

Documentation

• DEPLOYMENT.md – Deployment guide
• VALIDATION.md – Problem validation and track fit

Made with ❤️ by Team Yin & Yang for Protex Hack-2-Win 2026