Echo Bay - Murder Mystery

main startup screen
maya's notification screen
maya's lock screen
maya's home screen
maya's gallery
maya's calculator
tide level app
sonar app
evidence board
settings screen
main gallery app
phone app: recents
phone app: voicemail
phone app:contacts
phone app: keypad
helix corp page
chat/messaging app
browser app
DM/ messaging screen
phone calling screen
Spectogram mini game
Photo restoration mini game
maya's notification screen

Inspiration

Most murder mystery games are heavily scripted. Once you know the solution, the magic is gone. We wanted to build a mystery that feels alive, where suspects remember what you say, react emotionally, and change their behavior based on how you investigate. Our goal with Echo Bay was to create a case that unfolds through conversation, pressure, and choice, not prewritten dialogue trees.

How we used Gemini 3

Echo Bay is powered entirely by Gemini 3. We use its 1M-token context window to maintain full narrative continuity across the entire investigation, including clues, timelines, alibis, and character emotions. Every suspect remembers past conversations, what evidence has been revealed, and how the player has treated them.

Gemini 3 handles the reasoning behind branching narratives while keeping the mystery logically consistent and solvable. We use structured prompts and constraints to lock in the core plot while allowing open-ended dialogue and unexpected player choices.

For real-time interaction, we integrated the Gemini Live API with gemini-live-2.5-flash-native-audio, allowing players to speak naturally, interrupt suspects, and apply pressure. Responses are fast, expressive, and sensitive to tone.

How we built it

We built Echo Bay as a persistent AI-driven world. Gemini manages memory, reasoning, and character behavior, while our orchestration layer tracks evidence and investigation state and feeds structured context back into the model. Additionally, we used Firestore to store chat messages and story state in real time, and Flutter to deliver a high-fidelity, mobile-first frontend.

Challenges we ran into

The hardest part was balancing freedom and structure. We wanted players to say anything, but the mystery still had to make sense. Keeping suspects emotionally realistic without breaking facts required careful prompt design and validation.

Accomplishments that we're proud of

A fully voice-driven murder mystery
Persistent memory across an entire case
Suspects that adapt to player behavior
A mystery that is replayable and coherent

What we learned

Large-context models like Gemini 3 enable a new kind of interactive storytelling where AI is not just responding, but actively maintaining a believable world.

What's next for Echo Bay - Murder Mystery

We plan to expand Echo Bay with larger cases, deeper character relationships, and longer investigations that persist over time, all powered by Gemini 3.