Podit — AI Voice Event Agent
A hybrid voice + text agent that plans, schedules, and protects your calendar
Podit is a hybrid voice and text AI agent for intelligent event planning and scheduling. Users talk or type to plan events; the agent detects scheduling conflicts, books into Google Calendar, and respects personal guardrails like sleep and work hours so it never schedules over them. The voice loop runs at sub-500ms latency, fast enough to feel like a real conversation rather than a delayed assistant.
The pain this had to solve
Scheduling is one of the most common things people want to hand to an assistant, but most voice agents fail at exactly the part that matters: they book over conflicts, ignore personal boundaries like sleep and work hours, and lag badly enough that the conversation feels broken. A scheduling agent that double-books or wakes you up is worse than no agent at all.
Podit needed to be conversational and fast, but also disciplined — it had to reason over a live calendar, catch conflicts before committing, and honor each user’s preferences as hard constraints, all while staying responsive enough on voice to feel natural.
What I built — the architecture
One agent serves both a voice channel (via Twilio) and a text channel, so users can switch between speaking and typing without losing context.
Tuned the speech-to-response pipeline — streaming, model selection, and pruned tool calls — to keep round-trip latency under 500ms so the conversation feels live.
Before booking, the agent reads the live Google Calendar and checks for overlaps, flagging conflicts and proposing alternatives instead of double-booking.
User preferences such as sleep and work hours are enforced as dynamic guardrails, so the agent never schedules into protected time even when asked carelessly.
A LangGraph state graph (on OpenAI models) sequences understanding, calendar reads, conflict checks, and confirmation, with shared state persisted in Supabase and a React Native client.
What it delivered
Podit holds real-time voice conversations at sub-500ms latency, reads and writes a live Google Calendar, catches conflicts before booking, and enforces sleep/work-hour guardrails so it never double-books or schedules over protected time — a scheduling assistant users can actually trust with their calendar.
“A dependable engineer who can be trusted with complex, high-stakes work”
— Ajay S., Founder
Want results like this for your team?
Tell me what you want to automate. On a free 30-minute call I’ll tell you straight whether it’s worth building, roughly what it costs, and how I’d approach it — no pitch, no obligation.
Book my free 30-min AI scoping call →