Life Rewind

Agent Reasoning

🌟 Inspiration

Digital truth is rarely clean. In real incidents, evidence is scattered across screenshots, chat logs, emails, videos, and metadata—often incomplete, unordered, or contradictory. Reconstructing what actually happened requires time, expertise, and careful judgment. We built Life Rewind to shift this burden from humans to an autonomous AI agent.

🧠 What it does

Life Rewind is a Gemini 3–powered autonomous AI agent that reconstructs what really happened from fragmented digital evidence. Users upload multimodal inputs such as screenshots, chat exports, and files. The Life Rewind Agent independently correlates timelines, detects inconsistencies, identifies missing evidence windows, and asks targeted follow-up questions. It continuously re-evaluates the narrative as new evidence is added and generates a confidence-aware, export-ready investigation report.

🛠️ How we built it

Life Rewind is built around an agentic reasoning pipeline powered by the Gemini 3 API. Gemini 3’s multimodal inference is used to ingest and understand mixed evidence types, while its long-context reasoning enables cross-document timeline reconstruction. An orchestrator layer manages autonomous cycles—parsing evidence, detecting gaps, re-running reasoning, and assigning confidence levels. The frontend visualizes agent reasoning steps, timelines, and evidence correlations in real time.

⚠️ Challenges we ran into

The biggest challenge was avoiding overconfidence. Real-world evidence is messy, and forcing a single “clean” answer risks hallucination. We addressed this by designing the agent to explicitly surface uncertainty, flag weak inferences, and re-run its reasoning when new evidence appears. Balancing autonomy with transparency was critical.

🏆 Accomplishments that we’re proud of

We built a system that behaves like an investigator, not a chatbot. Life Rewind autonomously detects missing evidence, asks for clarification, revises its conclusions, and exposes its reasoning process on screen. The confidence-aware timeline and structured report demonstrate real-world usefulness beyond simple analysis.

📚 What we learned

Agentic systems feel fundamentally different from prompt-based tools. Showing autonomy, self-doubt, and re-evaluation builds trust far more than confident outputs alone. Gemini 3’s multimodal reasoning and long-context capabilities are especially powerful when paired with an explicit agent loop rather than a single prompt.

🚀 What’s next for Life Rewind

Next, we plan to deepen agent autonomy by enabling proactive evidence requests, richer confidence metrics, and optional integrations with communication platforms. We also aim to expand real-time reasoning and collaborative investigation workflows, turning Life Rewind into core infrastructure for digital truth reconstruction.