Multi-agent AI architecture in partnership with The Jackal. Our AI editorial room observes, analyzes, discusses and judges live performances, producing 5 artist ratings every evening for Sanremo 2026.
LangGraph-orchestrated AI agents working in parallel to transform live events into publishable content.
Critic, Press, and Tabloid agents analyze the same event through different editorial lenses.
Deterministic workflow with parallel processing. Agents work simultaneously then converge.
Ratings, reviews, dialogues, and social carousels generated automatically.
In partnership con The Jackal per Sanremo 2026. Genera 5 pagelle di artisti ogni sera pubblicate su Facebook, oltre a discussioni approfondite dalla nostra piattaforma multi-agente.
As the artist hits the stage, the system ingests lyrics, video, images, and archive documents into structured evidence packages.
Three agents work simultaneously on the same data, each with different voice, age, and perspective.
The Moderator builds dialogue between desks. Output: reviews, carousels, and social content.
Sanremo AI Lab
Each evening of the Festival, every performance generates a wave of content, comments, memes and rankings. At Btinkeeng, we wanted to contribute to this debate in our own way, building an entire editorial team made of artificial agents.
Contribuire al dibattito con la nostra prospettiva — costruendo un'intera redazione fatta di agenti artificiali.
Un sistema multi-agente basato su AI generativa — una macchina narrativa che produce output multi-formato.
Per ogni artista in gara, un team di agenti AI specializzati — ciascuno con voce, età e prospettiva diverse — analizza la performance e produce contenuti pronti per la pubblicazione.
The moment an artist steps on stage, the system has already begun processing everything into a structured package of evidence.
LangGraph orchestration layer - Parallel agent processing workflow
The heart of the system is a graph of interactions for each episode. The flow is deterministic but parallel in the central phase:
Data is analyzed and constitutes the context
Three specialized agents then operate in parallel on the same artist
Simultaneous execution reduces times and preserves editorial independence
Each agent has an LLM model assigned based on the task. The music critic uses a high reasoning capacity model; the tabloid favors fluidity and lively tone; the press agent works with a model optimized for structured outputs.
All models are called through a unified external LLM gateway
After processing individual episodes, a second level of orchestration aggregates the results together. A Pagellista agent produces evaluation sheets for each artist, while the Moderator agent builds the dialogue between agents and collects insights in the form of carousels for Instagram.
The system instantiates multiple autonomous agents working in parallel. Data flows from the three specialized desks through the Moderator and Pagellista, transforming into ready-to-publish editorial content.
Art and craft lens. Deep analysis of writing, interpretation, aesthetics.
Segnali radiofonici. Hook, rotazioni, potenziale virale.
Visual impact and pop culture. Outfits, gestures, memorable frames.
Brings voices together, builds dialogue, highlights contrasts.
Votes by category, rankings, special prizes.
Structured ratings with micro-reviews for each category: voice, lyrics, stage presence, emotion, hype.
TV talk show-style discussions synthesizing all three perspectives into long-form content.
L'architettura è progettata per essere ricalibrata su qualsiasi evento che genera un flusso di dati in tempo reale.
Richiedi DemoDa Sanremo al tuo prossimo evento. Sport, conferenze, lanci prodotto — architettura scalabile per ogni occasione live.