Building PRAAPTI AI: From SAM & Docker Friction to Zero-Trust Civic Intelligence with AWS Cedar and Strands
How our hackathon team navigated early containerization bottlenecks to build a 100% local, zero-trust civic intelligence platform for Indian welfare access and statutory RTI drafting using AWS Cedar, Strands Agents, and OpenSearch.
Every year, thousands of welfare schemes launched by central and state governments across India fail to reach the citizens who need them most. Whether it is a marginal farmer navigating crop subsidy ceilings, a rural student applying for post-matric scholarships, or an artisan seeking PM Vishwakarma equipment incentives, the barrier is rarely eligibility, it is information asymmetry, ambiguous documentation checklists, and administrative inertia. When applications stall indefinitely, ordinary citizens lack an accessible, legally sound mechanism to draft statutory Right to Information (RTI) inquiries to demand accountability.
During the WeMakeDevs AWS First Commit Hackathon (Track 1: Build It), our 4-person team set out to solve this by engineering PRAAPTI AI (प्राप्ति AI): Pradhanmantri and Rajya Assistance Application, Probability, and Transparency Interface.
Here is the honest story of our build: how we started on AWS Skill Builder, hit a wall with local containerization, pivoted to a clean local architecture, and vibe coded a 14-rule zero-trust civic intelligence system using AWS Cedar, Strands Agents SDK, and OpenSearch.
The Initial Roadblock: When Local Docker Emulation Consumed the Clock
We started our hackathon journey by diving into AWS Skill Builder digital courses and documentation to understand AWS open-source security and search tooling. Excited to build an enterprise-ready serverless blueprint, our initial instinct was to package everything using the AWS SAM CLI and Docker template.yaml.
However, within the first 12 hours, reality set in.
Setting up Docker daemon volume bindings on Windows, troubleshooting sluggish container cold starts, and debugging local Lambda runtime networking created immense friction for our frontend and API teammate, Sanjay. We found ourselves burning valuable hackathon hours fixing local container networking rather than refining our core civic intelligence algorithms.
At 5:00 at the PM (20th of September), 3 hours before the deadline ;-;, we made a crucial architectural call: drop SAM and Docker completely, strip out container layers, and build a unified, native local Python server architecture. That pivot freed up our momentum and allowed us to focus 100% on what mattered: intelligent agent workflows and zero-trust policy enforcement.
Decoupling Security: The Power of the AWS Cedar Policy Language
The most critical challenge in civic technology is authorization. While discovering welfare schemes must be universally open to all citizens, generating statutory RTI legal notices—especially Tier 3 Central Information Commission (CIC) Second Appeals under Section 19(3)—carries quasi-judicial weight. Allowing unverified accounts or automated bots to spam appellate tribunals could lead to severe systemic abuse.
Initially, handling these permission tiers resulted in brittle, nested if-else statements scattered across our Python codebase.
Discovering AWS Cedar (auth.cedar) was the biggest breakthrough of our hackathon. Cedar enabled us to decouple access control entirely from business logic into clean, declarative, human-readable zero-trust policies.
Our teammate Ratheeswar expanded this into a comprehensive 14-rule security matrix covering:
Statutory fee exemptions: Section 7(5) automatic waivers for BPL cardholders, senior citizens (60+), and Persons with Disabilities (PwD).
Life and Liberty Provisos: Section 7(1) 48-hour emergency RTI handling.
Abuse protection: Daily rate-limiting caps, anti-bot safeguards, and strict PII / unredacted Aadhaar data protection rules.
Dynamic Intelligence with Strands Agents and OpenSearch
With zero-trust authorization locked down, we structured the core recommendation pipeline around the AWS Strands Agents SDK and OpenSearch:
Multi-Parameter Knowledge Base (Harsh): We ingested 500+ lines of comprehensive welfare schemes across 15+ central ministries into OpenSearch, mapped alongside 700+ Indian districts. OpenSearch hybrid querying allows sub-millisecond filtering across income ceilings, land-holding acreages, and affirmative caste categories.
Dynamic Strands Orchestration (Sushanth): Decorated modular Python tool definitions to evaluate empirical approval odds, identify missing verification documents, and activate fallback closest recommendations when strict criteria are not met.
AWS GuardDuty and Route53 Threat Heuristics: Embedded live socket DNS resolution and sovereign registry checks (.gov.in, .nic.in) to scan suspicious URLs and warn citizens against advance UPI/QR registration scams.
Accessible Civic Dashboard (Sanjay): Built a high-contrast, responsive interface with real-time probability visualizers, dynamic document modals, and dark mode support.
Vibe Coding and Lessons Learned
Building PRAAPTI AI in 4 days was an intense experience in modern engineering. Pairing developer intuition with AI coding assistants like Antigravity, Gemini, and ChatGPT allowed us to debug AST parsing for Cedar native bindings, validate complex Pydantic schemas, and rapidly iterate on UI state.
Our key takeaway for builders: Do not let tooling complexity overshadow problem-solving. Recognizing when a tool (like local SAM container emulation) is slowing your team down and having the courage to pivot toward lightweight, sovereign open-source primitives (like Cedar and OpenSearch) is what transforms a stalled prototype into a working, high-impact product.
Architecture and Code Repository
Built by: Sushanth Arun (Strands Agent Core), Sanjay (Civic UI), Ratheeswar (Cedar Policies), and Harsh (OpenSearch Engine).
PS: In peak hackathon fashion, we started recording our demo video at 7:53 PM with seven minutes on the clock, watched in pure panic as YouTube’s video processing bar crept past the 8:00 PM cutoff, and were saved at the absolute wire when the magical deadline extension email landed in our inboxes, extended till midnight :-).
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article