LifePilot

💡 Inspiration

A few months ago, I audited my bank statement and discovered I had been paying for two streaming services I hadn’t opened in eight months, a gym membership I thought was frozen, and an annual software renewal that quietly doubled its price.

Most people don't struggle with big life decisions—we struggle with the death by a thousand micro-errands. The recurring bill that silently increases, the car maintenance schedule you forget until the dashboard light flashes, the forgotten receipt that was due for reimbursement. Existing apps either bombard us with dozens of useless push notifications or require hours of manual data entry.

We didn't need another noisy dashboard or another chatbot to have small talk with. We needed an assistant that behaves like an attentive personal manager: running silently in the background, handling the tedious busywork, and reaching out only when a genuine human decision is required. That became the vision for LifePilot.

🚀 What It Does

LifePilot is an autonomous background everyday agent powered by the Strands Agents SDK. It continuously manages the background friction of daily living across three core domains:

1. Subscription & Bill Guardian: Monitors recurring digital invoices, flags unannounced price hikes, identifies duplicate or dormant subscriptions, and drafts cancellation or renegotiation requests.
2. Household Logistics & Errand Dispatch: Tracks routine domestic maintenance (filter replacements, vehicle inspections, warranty expirations) and pre-organizes the quotes or order links.
3. Calm "One-Touch" Human Gate: Instead of badgering you with alerts, LifePilot bundles pending items into a daily or weekly executive digest with explicit, one-click actions: [Approve Cancellation], [Authorize Order], or [Dismiss].

🛠️ How We Built It

We built LifePilot from the ground up around the philosophy that the best agents are mostly invisible:

• Strands Agents SDK: Forms the brain of LifePilot. We utilized Strands to orchestrate autonomous multi-step reasoning chains, stateful memory persistence, and dynamic tool invocation (calendar integration, email/invoice parsing, and web search for better service deals).
• Amazon Bedrock & AgentCore: Provides the foundational LLM intelligence for nuanced document understanding and strict JSON tool calling, ensuring sensitive financial and scheduling data is parsed with near-zero error.
• Python / Fast-API Engine: Handles asynchronous task scheduling, event webhooks, and secure local credential caching.
• Calm Interface: Designed an uncluttered, distraction-free web dashboard focused purely on action queues rather than cluttered graphs.

🚧 Challenges We Faced

1. The "Notification Paradox": Teaching an autonomous agent when NOT to talk is significantly harder than getting it to speak. Early iterations pinged too frequently about trivial updates. We implemented a strict confidence-and-impact scoring heuristic: if an action is non-destructive and saves under \$5, it's logged silently; if it requires spending money or canceling a contract, it triggers the Human-in-the-Loop approval gate.
2. Accurate Document & Receipt Parsing: Receipts and bill statements come in wildly inconsistent layouts. We leveraged few-shot tool prompting within the Strands framework to ensure mathematical totals and dates were verified deterministically before the agent acted on them.
3. State Management across Multi-day Tasks: A cancellation request or service quote often takes 24–48 hours to resolve. Implementing robust conversational checkpointing in Strands allowed LifePilot to pause execution safely and resume without losing context.

🏆 Accomplishments We're Proud Of

• True End-to-End Autonomy: LifePilot doesn't just suggest actions—it prepares the complete execution payload (draft email, prepopulated cart, or calendar block) so the user only has to approve.
• Zero Hallucination Tolerance: Built deterministic verification guardrails around all monetary and scheduling calculations.
• A Product That Feels Peaceful: Creating an AI experience that genuinely gives you time back rather than demanding more screen time.

📚 What We Learned

• Building agentic software requires a mindset shift from reactive prompting (chatbots) to proactive event loops (agents).
• Human-in-the-loop UX is the true key to user trust. People love delegation, but they demand veto power.
• The Strands Agents SDK drastically simplifies the complexity of coordinating state, tools, and LLM backends into clean, maintainable modular steps.

🔮 What's Next for LifePilot

• Proactive Bill Negotiation: Equipping LifePilot with automated call/chat scripts to renegotiate broadband and insurance rates.
• Family & Household Sync: Enabling multi-user shared approval queues for couples and family caretakers.
• Voice & Messaging Integration: Bringing LifePilot’s approvals directly to WhatsApp, Telegram, or SMS for frictionless on-the-go confirmations.

Built With

• agentic-workflows
• ai-agents
• amazon-bedrock
• amazon-web-services
• automation
• autonomous-agents
• css3
• fastapi
• html5
• human-in-the-loop
• javascript
• llm
• productivity
• python
• rest-api
• strands-agents-sdk