Skip to content

VALIDATION.md

Latest commit

 

History

History
72 lines (45 loc) · 4.7 KB

VALIDATION.md

File metadata and controls

72 lines (45 loc) · 4.7 KB
Raw

VALIDATION - Redressa

Short validation notes for hackathon pitching and PPT use.

The Problem Is Real

  1. Consumers get stuck after first-line support fails. India's consumer forums received over 900,000 complaints between 2020-2023 (National Consumer Disputes Redressal Commission data). Most consumers give up before that stage because they don't know how to escalate - they lack the right language, the right contact, and the right evidence structure.

  2. Policies and escalation paths are buried and fragmented. Airline passenger rights (DGCA CAR Section 3, Series M Part IV), e-commerce return policies, and consumer protection rules (Consumer Protection Act 2019, E-Commerce Rules 2020) exist - but they're scattered across PDFs, policy pages, and regulatory notices that no frustrated consumer will read at the point of complaint.

  3. No existing tool bridges the gap between raw complaint and structured escalation. Generic chatbots give surface-level advice. Legal platforms target formal litigation. Nothing currently takes a consumer's messy evidence (screenshots, invoices, chat logs) and produces a grounded, policy-cited, escalation-ready claim package.

Why This Fits Track 2 - Agentic AI / AI Workflows

Track 2 asks for: autonomous agents capable of multi-step reasoning and real-world task execution.

Redressa is exactly this:

Track 2 Requirement How Redressa Delivers
Multi-step reasoning 9-step typed pipeline: intake -> parse -> extract -> timeline -> classify -> retrieve policy -> evaluate -> route -> generate
Autonomous agent behavior Each step runs as a visible agent action with structured outputs - no manual intervention between steps
Real-world task execution Produces actionable outputs: grievance emails, escalation notes, evidence checklists - not just analysis
Visible agent workflow Agent Activity panel shows step-by-step progress with timestamps and citations

This is not a chatbot wrapper around an LLM. It's a workflow agent that does multi-step reasoning over real documents and produces real-world-usable outputs.

Why These Two Complaint Categories First

IndiGo flight cancellation / refund disputes

  • High volume: IndiGo is India's largest domestic airline (~60% market share)
  • Clear regulatory backing: DGCA passenger rights rules define compensation obligations
  • Structured evidence: PNRs, booking confirmations, and cancellation emails are machine-parseable
  • Emotionally resonant demo: everyone has experienced a flight disruption

Flipkart damaged / defective goods

  • High volume: Flipkart is one of India's two largest e-commerce platforms
  • Clear policy structure: published return/refund/replacement policies
  • Common pain point: damaged or wrong item delivery is one of the most frequent consumer complaints
  • Easy demo evidence: order confirmations, delivery photos, chat transcripts

Both categories share these properties:

  • Well-documented company policies exist
  • External regulatory guidance applies
  • The evidence types are structured enough for deterministic parsing in an MVP
  • The escalation paths are concrete (grievance cell, nodal officer, DGCA, consumer forum)

Starting narrow with two strong categories is better than starting broad with ten weak ones.

Why the Phased Approach Is Right for a Beginner Team

  1. Each phase is independently demoable. If time runs out after Phase 1, the project is still a valid, complete hackathon submission. There is no "all-or-nothing" risk.

  2. Backend-first ordering prevents the fake-demo trap. Building schema -> pipeline -> UI (in that order) means the demo shows real data flowing through real logic, not hardcoded mock screens. Judges will ask how it actually works.

  3. Scope discipline protects build velocity. Features like browser automation, live scraping, vector search, and outbound email are explicitly deferred to Phase 3. This prevents the team from chasing impressive-sounding features that are fragile, hard to debug, and likely to break during a live demo.

  4. The priority order is risk-aware. The build plan deploys early (step 1.4) instead of at the end, seeds demo data before polish, and keeps one stable demo path alive at all times - all patterns that reduce the chance of a last-minute scramble.

Summary for Pitching

"Every year, lakhs of Indian consumers give up on valid complaints because they can't navigate the escalation maze. Redressa is an agentic workflow that takes their messy evidence - screenshots, invoices, chat logs - and produces a grounded, policy-cited, escalation-ready claim package in minutes. It's not a chatbot. It's a consumer redressal agent that actually does the work."