LifeShield Everyday Human Agent

Turn everyday corporate fine-print traps into citizen victories with an autonomous statutory defense agent built on AWS Bedrock and Strands Agents.

Comment

🛡️ LifeShield Core: Autonomous Everyday Human Rights & Dispute Defense Agent

AWS Agents for Humans Hackathon 2026 — Everyday Agents Track
Turning the tables on corporate fine-print traps with the Strands Agents SDK and Amazon Bedrock.

💡 Inspiration

Every single day, ordinary citizens fall victim to everyday corporate fine-print traps:

  • Landlords arbitrarily withholding ₹40,000 / $500 security deposits citing "routine repainting" or normal wear-and-tear.
  • Airlines cancelling flights at the last minute and refusing full cash refunds, pushing useless expiring vouchers instead.
  • Banks & FinTechs levying unauthorized transaction charges and hidden debit fees.
  • Dark Patterns & Subscriptions locking users into recurring debits with hidden cancellation buttons.

Large corporations have dedicated legal departments, boilerplate disclaimers, and automated rejection bots. Ordinary citizens have zero defense — hiring a lawyer is prohibitively expensive, and navigating legal portals is intimidating. Most people simply give up and lose their hard-earned money.

We asked: What if an everyday human had an autonomous, tireless legal defense agent in their pocket?
An agent that doesn't just chat, but actively diagnoses violations, cites codified government statutes across 195+ countries, and compiles court-admissible 7-day demand notices in under 2 seconds. That vision became LifeShield.

🛠️ How We Built It

LifeShield is architected as an autonomous multi-stage agent pipeline combining the Strands Agents SDK and Amazon Bedrock:

1. Strands Agents SDK Core

We leveraged the Strands Agents SDK to build modular, deterministic @tool functions that coordinate the defense lifecycle:

  • DiagnosisTool: Performs rule-based and semantic classification of citizen grievances across 20+ dispute categories (Tenancy, Banking, Aviation, E-commerce, Telecom, etc.), scoring violation severity and bad-faith conduct.
  • LegalStatuteTool: Retrieves codified statutory protections (such as the Consumer Protection Act 2019, Model Tenancy Act, DGCA Civil Aviation Requirements, and RBI Master Directions) with zero hallucination.
  • DocumentTool: Generates formal, court-admissible 7-Day Statutory Notice PDFs with unique case docket IDs, barcode headers, legal notice clauses, and formal signature blocks.
  • LifeShieldAgent: Orchestrates these tools within a native Strands agent lifecycle, tracking execution latency, tool success states, and evidentiary outputs.

2. Amazon Bedrock & Strategic Reasoning

We integrated Amazon Bedrock (using the modern Universal Converse API with Amazon Nova / Anthropic Claude 3 Haiku) to synthesize strategic negotiation counsel, dispute leverage analysis, and ombudsman escalation pathways tailored to the user's jurisdiction.

3. Evidentiary Chain-of-Custody & Memory

  • Immutable Case Docket: Backed by SQLite, all cases, generated PDFs, and audit logs are recorded with timestamped cryptographic docket IDs.
  • Multi-Turn Conversational Memory: The agent remembers case details across conversation turns, allowing users to submit new evidence, clarify facts, or adjust claim amounts dynamically.

4. Modern, Accessible Citizen UI

Built with Streamlit, custom glassmorphic styling, and typography from Google Fonts (Outfit & Plus Jakarta Sans), the interface supports 195+ sovereign nations, 6+ languages (English, Tamil, Hindi, Spanish, French, German), and preloaded benchmark disputes for instant demonstration.

⚡ Challenges We Faced

  1. Deterministic Law vs. Probabilistic LLMs:
    Legal citations and penal clauses cannot tolerate hallucination. Relying solely on raw LLM prompts often produced fabricated section numbers. We solved this by separating concerns: deterministic Strands Agent tools handle statutory verification and docket IDs, while Amazon Bedrock handles strategic legal advice and communication tone.

  2. Multi-Jurisdiction Statutory Complexity:
    Consumer rights vary drastically between jurisdictions (e.g., US FTC regulations vs. India's CPA 2019 vs. UK Consumer Rights Act). We codified statutory rule trees across 195+ countries so the agent seamlessly adapts governing laws based on the user's selected jurisdiction.

  3. Sub-2-Second End-to-End Latency:
    Compiling legal PDFs, querying Bedrock, and diagnosing violations can become slow. By streamlining our tool pipeline and leveraging lightweight models like Amazon Nova Micro and Claude 3 Haiku, we achieved an end-to-end execution time of ~1.2 seconds.

  4. Zero-Friction Citizen Experience:
    Legal forms are notoriously overwhelming. We designed the interface with smart pre-filled defaults, 1-click sample loaders, and automated email dispatch so everyday users never have to write legal jargon themselves.

🎓 What We Learned

  • The Power of Agentic Tool Calling: Combining the Strands Agents SDK with Bedrock demonstrated how agents become genuinely useful when given real-world actuators (PDF generation, database logging, email drafting) instead of remaining mere chatbots.
  • Citizen Empowerment through AI: Complex corporate terms of service are designed to exhaust consumers. AI agents can democratize legal knowledge, giving ordinary citizens the same institutional leverage as large corporate legal departments.
  • Resilient Fallback Architectures: Building systems that remain functional across cloud-connected and offline environments ensures reliability when citizens need defense the most.

🚀 What's Next for LifeShield

  • WhatsApp & SMS Voice Gateway: Allowing non-technical citizens to send a voice note on WhatsApp and receive a court-ready PDF notice directly on their phone.
  • Automated Ombudsman API Filing: Direct API integration with government dispute portals (e.g., National Consumer Helpline, RBI CMS portal, DGCA AirSewa).
  • Collective Bargaining & Class-Action Clustering: Grouping similar grievances against the same corporate entity to initiate collective multi-citizen claims.

Built With

Share this project:

Updates

Submission history