Glean on Alexa+ | Family Paperwork Assistant

Glean is a proactive voice assistant that protects multigenerational families by tracking vital documents, unlocking welfare schemes, and simplifying bureaucracy across every life stage.

Comment

Inspiration

In every multi-generational household, family administration is an invisible, high-stakes mental burden. Usually, one family member shoulders the responsibility of tracking passports, health insurance cards, vehicle registrations, pensions, and statutory filings across grandparents, working parents, and students—each with conflicting expiration dates and zero centralized tracking.

Critical documents expire silently, leading to emergency renewal fees, delayed vacations, or lapsed healthcare coverage. Even worse, governments offer life-changing statutory schemes (like Medicare, Senior Citizens Savings Scheme, or Ayushman Bharat PM-JAY), yet millions miss out because eligibility criteria are buried in dense bureaucratic text.

We asked ourselves: What if Alexa wasn’t just a passive assistant waiting for commands, but an empathetic, proactive family guardian that foresees administrative deadlocks before they happen and explains complex paperwork in language every generation understands? That was the spark for Glean.

What it does

Glean is a proactive family administrative agent built for the Alexa+ ecosystem over the Model Context Protocol (MCP):

Continuous Family Document Sentinel: Audits document validity across all household members, tracking expiration countdowns and alerting the family with proactive urgency rankings before critical paperwork lapses.

Dual-Jurisdiction Welfare Engine (US & India): Evaluates multi-generational profiles against federal and statutory welfare programs (Medicare, SNAP, Ayushman Vay Vandana, SCSS, APY, PMJJBY), instantly matching eligible schemes and flagging required verification steps.

36-Month "Almost Eligible" Forecasting: Identifies family members approaching milestone statutory birthdays (e.g., reaching 65 for Medicare or 70 for Ayushman Vay Vandana) up to 36 months ahead, queueing paperwork preparations well before the milestone arrives.

Document Prerequisite & Critical Path Auditing: Cross-references family vault documents against statutory scheme requirements to detect blocker cascades (e.g., an expired certificate or missing proof blocking a high-value benefit), mapping out an optimal step-by-step resolution order before in-person appointments.

Bureaucratic Conflict Simulator: Simulates red tape and conflicting statutory rules (e.g., dual-enrollment restrictions between private insurance and state benefits, or girl-child quota caps) to alert families to hidden disqualifications and suggest remedial restructuring steps.

Autonomous Life Event Orchestration: Constructs phased, actionable 90-day roadmaps for milestone transitions including Retirement, Adulthood Independence, and Inter-State Relocation.

Cross-Generational Voice Dialect Engine: Dynamically reframes dense governmental text into the right conversational tone—reassuring, zero-jargon analogies for grandparents, structured fiscal alpha for students, and tax-focused efficiency for working parents.

Proactive Ambient Push (SSE): Emits real-time Server-Sent Events to push urgent alerts to the Alexa screen and audio chime when action is needed, without waiting for the user to ask.

How I built it

Backend & MCP Layer: Developed in Python 3.12 with the official Model Context Protocol (MCP) SDK over Streamable HTTP and SSE, featuring 17 discrete administrative tools, 4 resources, and 3 prompt templates.

Frontend & Simulator: Built using Svelte 5 and Vite, featuring an interactive central glowing voice orb, acoustic waveform visualizer, ambient status chips, and structured visual cards.

Speech & Audio: Integrated the Web Speech API with conversational text cleaners and Alexa audio chime integration.

DevOps & CI/CD: Continuous Integration pipeline with GitHub Actions running 261 automated tests and automated CD deployment to Render.

Data & Rule Engines: SQLite relational database paired with custom deterministic evaluation engines:

Prerequisite Cascade Evaluator: Maps vault documents to scheme prerequisites using semantic category aliasing (Identity, Address, Age, Bank), detecting expired blockers and near-term expiry risks.

Bureaucratic Conflict Detector: Rules engine checking statutory exclusivity clauses, income thresholds, and multi-scheme conflicts.

Readability & Persona Engine: Flesch-Kincaid grade analysis and jargon scoring to dynamically adapt voice output for each generation.

Challenges I ran into

Multi-Prerequisite Blocker Cascades: Modeling real-world paperwork bottlenecks required mapping complex, multi-document requirements to disparate vault entries with varying expiration dates, ensuring the agent prioritizes the single root blocker that unlocks multiple downstream schemes.

Strict Zero-PII Security: We refused to store raw identity numbers (no Aadhaar, SSN, or passport numbers in the database). Designing deterministic eligibility and tracking logic while strictly using document types, expiry dates, and relationship tags required careful architectural design.

Tone Adaptation Across Generations: Translating statutory legal codes into natural spoken Alexa answers without losing legal accuracy or sounding condescending required careful prompt engineering and readability scoring.

Push-Channel Synchronization: Establishing a low-latency Server-Sent Events (SSE) push channel that seamlessly wakes the ambient display card and audio chime when urgent deadlines occur.

Accomplishments

261 Automated Tests (100% Pass Rate): Engineered an exhaustive Pytest suite covering every edge case—expired dates, circular dependencies, edge-case milestone leap years, and cross-border statutory rules.

True Proactive Behavior: Moving beyond "Alexa, set a timer" to an assistant that speaks up when Amy’s pension certificate is 18 days away from lapse.

Cross-Generational Accessibility: Grandparents receive warm, stress-free voice summaries ("tea and fruit pocket money"), while students receive analytical summaries ("sovereign high-yield yield with zero drawdown").

Complete End-to-End Deployed Solution: A production-ready MCP server hosted on Render, compatible with the official MCP Inspector, Claude Desktop, and our rich Alexa Simulator.

What I learned

The Power of the Model Context Protocol (MCP): Treating capabilities as structured, discoverable MCP tools allowed us to build cleanly modular agents that reason over complex, multi-table household graphs without monolithic codebases.

Families Require Longitudinal Context: Real administrative assistants cannot forget context between sessions; long-term milestones (graduations, retirements, senior transitions) require multi-year timeline forecasting.

Spec-Driven Development Saves Time: Clear upfront contracts for tool inputs and schemas prevented interface drift between our Python backend, Svelte simulator, and voice client.

What's next for Glean on Alexa+ | Family Paperwork Assistant

Physical Alexa Hardware Integration: Deploying Glean as a native certified Alexa Skill with Alexa Presentation Language (APL) for Echo Show 8, 10, and 15 devices.

Zero-Knowledge Vision Document Auditing: Using Alexa’s camera or mobile companion app to snap a photo of an official renewal notice, extracting expiration dates and document types via local on-device OCR without cloud storage of images.

Automated Government Appointment Dispatching: Direct integration with public scheduling APIs (such as passport seva portals or DMV appointment systems) to hold slots when prerequisite documents are cleared.

Extended Civic Schemes: Expanding statutory databases to cover Canada, the UK, and European family welfare frameworks.

Built With

Share this project:

Updates

Submission history