AIdeas: Global Impact AI Navigator

My Vision

The Problem

What I Built

1. A SWOT diagnostic — sector classification, strength/weakness analysis, feasibility and impact scoring (0–100), SDG alignment, key risks, and actionable recommendations.
2. A live-funded opportunity portfolio — real grants, accelerators, competitions, and impact investment programs discovered via Amazon Nova Web Grounding (live internet search with citations), not hallucinated links. Users choose what they need (financing, mentoring, incubation, networking, training, technical support, market access, legal support) and the search is tailored to those needs.
3. A personalized roadmap — a time-boxed action plan (2 weeks → 6 months) tailored to the project's maturity, region, and SWOT findings, with specific funding recommendations and success factors.
4. A context-aware AI advisor — a chat interface that remembers the full project context (SWOT, opportunities, roadmap) and can answer follow-up questions, refine strategies, or brainstorm pivots.
5. A guided assessment quiz — for users who don't know where to start, a 5-question interactive quiz builds a structured idea and launches the full analysis pipeline automatically.

• Accessible — works on mobile, tablet, or desktop; works without AWS credentials in mock-data mode for demos
• Honest — scoring is conservative, weaknesses are surfaced alongside strengths, funding links are real (web-grounded)
• Actionable — every output is a concrete next step, not vague encouragement

Why This Matters

The information gap is real — and it costs the world

• Strategic clarity — Is my idea viable? What are the risks? What sector does it really belong to?
• Funding visibility — What grants exist for my specific sector, region, and maturity level? What are the deadlines?
• Execution guidance — What should I do in the next 2 weeks? 3 months? 6 months?
• Ongoing mentorship — I have follow-up questions, but I can't afford a consultant.

Why Social Impact matters here

How I Built This

Architecture Overview

AWS Services Stack

The Multi-Agent Pipeline — Strands Agents Deep Dive

Agent 1: The Diagnostic Agent (Nova Micro)

• The agent receives a detailed system prompt instructing it to classify the project by sector (10 categories), region (8 regions), and maturity (4 levels).
• It performs a SWOT analysis: 2–4 items each for Strengths, Weaknesses, Opportunities, and Threats.
• It scores Feasibility (0–100) and Impact (0–100) using a structured scoring rubric.
• It identifies 1–3 relevant UN SDGs.
• It surfaces 2–4 key risks and 2–4 actionable recommendations.

Agent 2: The Scout Agent (Nova 2 Lite + Web Grounding)

1. The agent uses Amazon Nova's 2 native Web Grounding — a built-in systemTool called nova_grounding that enables the model to perform live web searches during inference.
2. The API call includes toolConfig: { tools: [{ systemTool: { name: "nova_grounding" } }] }.
3. Nova searches the internet, synthesizes results, and returns text interleaved with citationsContent blocks that contain source URLs.
4. The agent extracts both the JSON output and the citation URLs, then enriches any opportunities that have missing or generic links with fuzzy title matching.
5. Every link is verified with an HTTP HEAD request (with GET fallback for 405/403 responses). Broken links are replaced with Google search fallback URLs.
   A freshness filter removes opportunities with expired deadlines or stale year references.
   6. Graceful fallback: If Web Grounding fails (e.g., missing bedrock:InvokeTool permission), the agent retries without grounding using Nova Lite's training knowledge. If that also fails, it falls back to a Bedrock Agent runtime call.

• The agent receives the complete context: the idea text, the full SWOT analysis, and the formatted list of funding opportunities.
  The user's expressed needs (financing, mentoring, incubation, etc.) are injected into the prompt, and the agent is instructed to prioritize and specifically address each need with concrete action steps.
• It builds a structured roadmap with:
  • Immediate Actions (2 weeks) — validation, quick wins
  • Short-Term Goals (1–3 months) — momentum, early traction
  • Medium-Term Milestones (3–6 months) — scaling, growth
  • Recommended Opportunities — top 2–4 from the scout results with reasoning
  • Key Success Factors — critical factors and pitfalls to avoid
• The roadmap adapts to the maturity level: Concept-stage projects get validation-heavy plans; Growth-stage projects get scaling strategies.

Agent 4: The Chat Agent (Nova Lite)

• A sliding-window token management system keeps the prompt under 30,000 tokens by trimming older messages while preserving the system context.
• The full SWOT analysis, scores, SDGs, risks, recommendations, opportunities, and roadmap are injected into every chat prompt so the advisor is always contextually aware.
  The chat is available as soon as the diagnostic is complete — users don't have to wait for the full pipeline.
  Output: Contextual responses that reference specific findings from the analysis.
  Why Nova Lite? Provides strong conversational quality with lower latency and cost than Nova Pro, with 2,000 max tokens and temperature 0.5.

Security & Production Readiness

AWS Free Tier Compliance

Frontend Features

• React 19 with lazy-loaded code-split routes (30+ pages accross projects)
• Tailwind CSS v4 with dark mode and glass morphism effects
• Framer Motion animations for page transitions and UI elements
• shadcn/ui component library for consistent, accessible UI
• Interactive SWOT Dashboard — color-coded 4-quadrant grid (Strengths in green, Weaknesses in amber, Opportunities in blue, Threats in red)
• Score Bars — animated progress bars with color coding (green ≥70, amber ≥40, red <40)
• SDG Badges — visual tags for each aligned Sustainable Development Goal
• Responsive Design — fully adaptive from 320px mobile to 4K desktop
• Progressive Loading — real-time status updates as each agent completes

Development with Kiro

Key Development Milestones

1. Spec & architecture with Kiro — Natural language requirements → structured specs → task breakdown
2. Project foundation — Express 5 + React 19 + TypeScript monorepo with shared types
3. Multi-agent orchestration — Sequential pipeline with error propagation and fallback paths
4. SWOT diagnostic overhaul — Expanded from 3 fields to 13+ fields with scoring rubric
5. Web Grounding integration — Native nova_grounding systemTool replacing fake search loops
6. Chat agent with token management — Sliding-window context with SWOT injection
7. Security hardening — Rate limiting, Zod validation, Helmet, CORS, input sanitization
8. Free Tier guardrails — 3-project limit, session management, optimized token usage
9. Assessment quiz — 5-question guided flow that auto-creates projects and launches analysis

Demo

1. Multi-agent orchestration is about context propagation, not just chaining

2. Web Grounding changes everything for trust

3. Kiro's spec-driven approach changes how you build AI systems

4. Prompt engineering is architecture, not copywriting

5. Graceful degradation is essential for production AI

• Scout Agent: Web Grounding → knowledge-only fallback → Bedrock Agent fallback
• All agents: robust JSON parsing that handles \<thinking\> tags, markdown code blocks, and extra text around the JSON
• Frontend: progressive loading with real-time status, error boundaries, and retry options
• Link verification: HTTP HEAD -> GET retry (for 405/403) -> Google search fallback for broken links

6. Free Tier design is a feature, not a constraint

• Use Nova Micro (cheapest) for structured classification, Nova Lite for search, Nova Pro (most capable) only for creative generation
• Limit projects to 3 per session to prevent runaway costs
• Keep token budgets tight (2,500 → 4,096 range) with temperature tuning
• Cache prompts in memory to avoid repeated disk reads
• Daily interaction limit of 2 per user for sustainable usage

7. SWOT analysis is the strategic backbone

• The scout became more targeted (searching for programs that match strengths and address weaknesses)
• The strategist became more actionable (building on strengths, mitigating threats)
• The chat agent became more contextually aware (answering questions with full SWOT context)

1. User agency improves output quality
   The decision to make the workflow user-guided — with a pause between diagnostic and opportunity search where users select their specific needs — was a critical design choice. Instead of guessing what the user wants, we ask. The needs selection (financing, mentoring, incubation, networking, training, technical support, market access, legal support) directly shapes:
   What the Scout searches for on the web
   How the Strategist prioritizes roadmap actions
   What kind of opportunities are surfaced first
   This simple interaction point dramatically improves relevance. A user who needs mentoring gets mentoring programs, not just grants. A user who needs legal support gets regulatory guidance programs, not just accelerators.

Diagrams & Visual Assets

Diagram 2: Web Grounding Data Flow

Diagram 4: Full System Architecture

Diagram 5: User Journey