Go-live: If you're stuck, just go-live

If you're stuck, just go-live

Inspiration
Home fixes hit at the worst times: Wi-Fi drops during work, breaker trips before dinner, kid locks the bathroom door, garage sensor fails on a rainy night in Strongsville. Most people end up lost in web searches or paying $100+ for a pro to arrive days later for a simple reset or reseat.

I wanted an AI that feels like a calm, local handyman right there with you—no waiting, no guesswork. Gemini Live’s real-time voice, vision, interruptions, and natural conversation made it possible: point your camera, talk like a friend, get safe step-by-step help instantly.

The name and slogan came naturally: Go-live: If you're stuck, just go-live.

What it does
Go-live is your real-time AI pocket handyman. Say “Hey Go-live” and point your phone camera at the issue—router, breaker panel, privacy lock, garage door, squeaky hinge, or similar basic home problem.

It analyzes live: LEDs, tripped breakers, pinhole locks, manual pages you show it. It guides naturally with short pauses (“Hmm… let’s see…”), empathy if frustrated, instant interruptions when you speak, micro-checks (“See the blue port?”), and strict safety stops (water near electricity, burning smell, exposed wires, nervous tone).

Most fixes happen in minutes: reseat cable, reset breaker once safely, unlock push-pin lock with paperclip, check garage alignment—no cost, no pro needed.

Go-live: If you're stuck, just go-live.

How we built it

Core: Gemini Live API (streaming voice in/out, real-time ~1 FPS vision, native barge-in, affective tone adaptation)
Frontend: TypeScript mobile/web app with live camera feed + cyberpunk AR overlays (cyan/amber highlights on ports, breakers, locks)
Flows: Structured steps per domain (router 5-step, breaker safety reset, lock pinhole, garage sensor) tracked in app state
Prompt: Single ordered system instruction — intent classification first, safety rule #1, natural fillers/pauses, manual reading via camera, domain pivots
Tools: Gemini function calling for structured vision (analyze_leds, check_breaker, read_manual_page)
Built fast using official Gemini Live examples + heavy prompt iteration for natural, safe feel

Challenges we ran into

Low-FPS vision reliability: LED/breaker/manual reads inconsistent in dim light → added explicit classification rules + user nudges (“Hold closer, better light”)
Robotic vs natural speech: Early “one sentence” rule felt stiff → relaxed to 1–3 sentences + fillers + empathy + questions
Mid-convo domain switches: Model sometimes stuck → fixed with clear intent-first classification + acknowledgment in prompt
Safety balance: Needed aggressive interrupt on risks without over-caution on safe tasks
Demo stability: Latency/Wi-Fi drops → prepped fallback clips + practiced recoveries

Accomplishments that we're proud of

Natural, human-like conversation: fillers, pauses, tone shifts, perfect barge-in — feels like talking to a real person
Multimodal personalization: Reads exact router/breaker/manual from camera, no uploads
Proactive safety shield: Interrupts on water, scorch, nervous tone, repeated trips
Multi-domain in one session: Router → breaker → lock → garage seamlessly
Full working prototype fast: live voice + vision + AR + structured flows end-to-end

What we learned

Prompt structure is very crucial: ordered rules, intent classification, safety priority, fillers make huge difference
Gemini Live’s barge-in, affective dialog, live vision are transformative; tiny additions (“Hmm…”, “Gotcha”) mask latency & build trust
Camera-pointing at manuals beats file uploads for natural feel
Safety must be rule #1 — proactive risk checks prevent misuse
Hackathons reward polish: vivid pain points, live interruption demos, sticky slogan turn good projects into standouts

What's next for Go-live: If you're stuck, just go-live

More domains: GFCI outlets, dryer vent lint, loose knobs, ceiling fan wobble, basic plumbing
Pro escalation: Book local electricians/plumbers/handymen in nearby via API
Offline basics: Pair with on-device Gemma for simple visual checks
Multi-language & accent support via Gemini’s 40+ languages
Ohio beta: Launch in Strongsville area, collect real-user feedback
Open-source prompt templates & flows to inspire safety-first home AI agents

Go-live turns “stuck” into “solved”—instantly, safely, confidently. Excited to keep building. 🚀