EventCast - AI Tinkerers Dublin: Agentic Architect Hackathon with Lyzr AI
AI Tinkerers - Dublin
Hackathon Showcase

EventCast

Team consisting of a Trinity College Dublin computer engineer and ML practitioner skilled in Python, PyTorch, and prototyping recommender systems and neural networks.

1 member

One-liner: A worldwide event-broadcast hub where a single post adapts and publishes to every event-hosting platform a user selects.

Usefulness — Solves a real problem people care about
Anyone running an event today posts the same thing 5–10 times across Eventbrite, Meetup, Luma, Facebook Events, Peatix, regional platforms, etc. Each platform has a different style, character limit, and audience expectation. EventCast turns that hour of copy-pasting into a 30-second flow: discover the right platforms for your region, pick the ones you want, write once, get platform-tuned variants ready to publish. The outcome is concrete — fewer hours wasted, broader reach, consistent brand voice across channels.

Autonomy — Agents do real, multi-step work
Two specialized agents do the thinking, not just answer a prompt:

Event Site Discovery Agent takes a location and returns a structured directory of relevant platforms with category, pricing model, audience size, supported event types, and a “why unique” line. It blends curated knowledge (30+ verified platforms) with live enrichment for regional specialists (Doorkeeper in Japan, Quicket in Africa, Sympla in Brazil, etc.).
Multi-Platform Post Composer Agent takes one master event description and produces a polished master post plus a uniquely adapted variant for each selected platform — respecting that platform’s voice (Meetup casual, Eventbrite SEO-structured, LinkedIn professional, Luma minimal), plus hashtags, image prompt, character counts, and per-platform status.
Both run end-to-end with no further prompting.

Clarity — Demo is obvious in under 30 seconds
The UI is a 3-stage horizontal journey shown in the sticky header: 01 Discover → 02 Compose → 03 Broadcast. A visitor lands on the hero, sees 30 platforms already loaded grouped by region, picks a few, fills a short event form, and watches the per-platform variants appear. Stage indicators, colored category dots, and a floating selection capsule make the flow self-explanatory — no tour needed.

Execution — Clean happy path, runs live
Input → process → output works end-to-end:

Curated results render instantly (zero API wait) so the page is never empty or broken.
The discovery agent enriches in the background with a 12-second timeout — if it fails, the curated list silently stays, so the user never hits a dead end.
The composer agent returns within seconds with the full structured response (master + per-platform versions + hashtags), and a simulated “Publish All” reveals the success state with per-card green confirmation.
Light + dark mode, mobile-responsive, reduced-motion respected, no flash of unstyled content.
Craft — Well-designed workflow, not a blob prompt
The workflow is properly decomposed:

Roles are separated — discovery and composition are independent agents with their own response schemas, not one mega-agent.
Tools — each agent is shaped by a strict JSON schema (strict: true, all fields required) so the UI consumes predictable data.
Steps — the platform list is a hybrid of deterministic curated data + agent enrichment, which makes the system robust to model failure.
Data flow — discovery output (selected platform names) is passed cleanly into the composer’s input. State stays consistent across the 3 stages.
Design system — 5-hue palette used functionally (one hue per category, one per stage), neuroaesthetic principles (golden-ratio columns, one accent per viewport, generous whitespace, soft motion curves), ambient mesh gradient backdrop for atmosphere without distraction.

Architect

deployed link of the app

Summarizing URL...