Clew Directive: A free personalized AI learning path

Clew Directive solves AI education overwhelm and FOMO. Take a 60-second quiz, get a personalized learning path from free resources powered by Amazon Nova reasoning. No paywalls. No tracking.

Comment

Inspiration

I was doomscrolling through Instagram one evening when I hit a breaking point. My algorithm had latched onto AI education, and suddenly every other post was another course, another "Learn AI in 10 Days," another promise. But here's what wasn't there: clarity. I couldn't tell which courses were actually free. Which were worth my time. Which one came next. I found myself knee-deep in research—tabs open, comparisons spreadsheets, analysis paralysis—just trying to figure out where to start.

And I realized: I wasn't alone in that overwhelm. Thousands of people are in that exact spot. They see the AI wave coming. They know they need to learn. But they don't know what AI even is, where the free resources are, or what path makes sense for them personally. The barrier isn't intelligence—it's information overload and inaccessibility.

That's when Clew Directive was born. Try it out at clewdirective.com

What it does

Clew Directive is a free, reasoning-powered learning path generator that cuts through the noise. Here's the flow:

  1. Answer 4 quick questions about your experience level, goals, learning style, and background (the "Vibe Check")
  2. AI synthesis — Amazon Nova's reasoning agents synthesize that into a personalized profile and confirm it with you
  3. Reasoning-driven path generation — Navigator Agent reasons through 28 curated, verified free resources to create a learning path just for you
  4. Download and learn — Get your personalized PDF. Start learning. No paywalls. No personal data collected. No accounts.

The core insight: AI-powered reasoning should democratize access, not gate it behind subscriptions.

How we built it

This was my first deep dive into production AI architecture, and I made deliberate choices around reasoning, privacy, and cost-efficiency:

Architecture: Multi-Agent Reasoning System

  • Scout Agent (Amazon Nova Micro): Independently reasons about resource freshness. Evaluates URL validity, content currency, pedagogical alignment, and global accessibility. Runs weekly via EventBridge, keeping the resource directory trustworthy without manual curation.
  • Navigator Agent (Amazon Nova 2 Lite): Synthesizes a user's 4-question profile into a coherent learning persona, then reasons step-by-step through 28 curated resources. Extended thinking and tunable reasoning effort enable genuine personalization—not template variations.

Technical Stack

  • Frontend: Next.js on AWS Amplify (deployed via console)
  • Backend: TypeScript CDK for infrastructure-as-code (Lambda, API Gateway, S3, EventBridge)
  • Deployment: Docker containers for Lambda functions (solves size limits and complex dependencies like WeasyPrint + Jinja2)
  • Monitoring: CloudWatch alarms and AWS Budgets to protect the $100 AWS credit allocation

Privacy by Design

  • Stateless architecture — no database, no user accounts, no personal data collected
  • PDFs auto-delete after 24 hours
  • Lambda logs expire after 7 days
  • Every design decision prioritized user privacy and trust

Cost Engineering

  • Nova Micro: 71x cheaper than comparable models, enabling continuous autonomous curation (~$0.00/month on Free Tier)
  • Nova 2 Lite: Fast reasoning model with tunable thinking effort — balancing instant response times with deep reasoning quality
  • Result: ~$11-12/month cost structure, delivering 6-8 months of runway on $100 AWS credit

Development Process

  • AI-assisted, human-reviewed: I used Kiro IDE, Claude, ChatGPT, and Gemini for scaffolding, code generation, testing, and architectural review
  • Every output was reviewed and tested
  • All decisions and mistakes are mine

Challenges we ran into

  1. Resource Verification at Scale — How do I verify 28 courses are actually good without manual curation? Built a 5-gate quality standard (authority, accessibility, currency, pedagogical soundness, global availability). The Scout Agent runs this pipeline weekly. It's not perfect, but it's honest and transparent.

  2. Personalization Without Tracking — How do I personalize without maintaining user databases or cross-session memory? Stateless design was the answer. The 4-question Vibe Check provides enough signal for Nova 2 Lite to synthesize a coherent profile and reason about path selection. Privacy preserved, personalization achieved, no personal data collected.

  3. Reasoning Quality vs. Speed — Nova 2 Lite is fast, but is it reasoning-quality for personalization? Extended thinking and tunable reasoning effort were key. Tested three personas (Skeptical Academic, Hands-on Builder, Business Professional) and confirmed the system generates genuinely different paths (9 unique resources across 3 users vs 4 identical in fallback). It works.

  4. Cost Constraints on Limited Budget — Can I build a production reasoning system on $100 AWS credit? Nova Micro is 71x cheaper than alternatives. Nova 2 Lite's price-to-reasoning ratio made the math work. Cost efficiency enabled impact, not diluted it.

  5. Lambda Deployment with Complex Dependencies — How do I deploy Lambda with WeasyPrint (C bindings) + Jinja2 + multiple Python dependencies? Docker container images for Lambda solved it—removed size limits, dependency hell, and cross-platform build issues.

Accomplishments that we're proud of

  • Deployed to production at https://clewdirective.com with live Nova reasoning pipeline
  • Multi-agent reasoning system that genuinely personalizes paths based on user profiles — not template-based recommendations
  • Reasoning-driven quality control — Scout Agent autonomously verifies resource freshness across 28 resources from 23 verified free courses
  • Production-grade architecture — Serverless, stateless, privacy-first design that costs ~$11-12/month
  • WCAG 2.1 AAA accessibility — 13.24:1 contrast ratios ensuring global usability
  • 9 unique resources across 3 test personas — proof that reasoning-driven personalization produces genuinely different paths, not variations of a template
  • Built with intention — Every architectural choice (privacy-first, cost-efficient, reasoning-powered) reflects the mission: democratizing AI education access

What we learned

  1. Reasoning is the differentiator, not speed or cost alone — Nova's reasoning capabilities enable genuine personalization. Cost efficiency doesn't dilute reasoning quality—it enables scalable impact.

  2. Stateless design unlocks privacy — You don't need databases or user tracking to deliver personalization. Stateless architecture + smart prompt design = privacy-first personalization.

  3. Infrastructure-as-code is non-negotiable for production — Building backend-as-code felt harder upfront (TypeScript CDK, Docker containers, deployment complexity), but it's production-grade, testable, and scalable. Worth the investment.

  4. AI-assisted development requires intentional review — I used multiple AI tools (Kiro, Claude, ChatGPT, Gemini) throughout development. Every output needs human review and testing. Full accountability for decisions and mistakes is non-negotiable.

  5. Technology is only good if it's accessible — Building fast was easy. Building fast and free and private and reasoning-powered? That required deliberate trade-offs. Accessibility (in cost, privacy, and usability) should be non-negotiable from day one.

What's next for Clew Directive

Short-term (Next 3-6 months)

  • Community partnerships with education nonprofits, bootcamps, and public libraries to embed Clew Directive into existing learning ecosystems
  • Email briefings (interface and HTML template ready, wire up Amazon SES)
  • Feedback loop: collect user outcomes to validate that personalized paths actually accelerate learning

Medium-term (6-12 months)

  • Multi-domain expansion: Schema supports domain field. MVP uses "ai-foundations." Future: AI for healthcare, educators, creatives, etc.
  • AgentCore Memory integration for optional cross-session context (with explicit user consent)
  • Community curation workflow via GitHub issues for resource contributions

Long-term (Vision) Clew Directive becomes the default "entry point" for new AI learners globally—a trusted, free, accessible resource that removes barriers and directs learners toward verified knowledge, regardless of geography, background, or budget.

Why this matters: AI literacy is becoming a baseline life skill. Access shouldn't depend on money, geography, or luck. Clew Directive removes those barriers. One personalized learning path. 60 seconds. Free forever.

Note on Development

This is an AI-assisted, human-reviewed build. All decisions and mistakes are mine. I used Kiro IDE, Claude, ChatGPT, Gemini, and other AI tools throughout development for scaffolding, code generation, testing, and architectural review. Every output was reviewed, tested, and integrated with full accountability for correctness and alignment.

Built With

Share this project:

Updates

Submission history