Casting Intelligence: A Mobile CRM Built on Relationships

The heart of the system. 119,000 persons and 220,000 working relationships from Swedish film and TV, structured as a graph where every node is a person and every edge is a collaboration. Edges are weighted by recency and role type — a director-actor collaboration last year weighs more than a co-production credit from a decade ago.

What the graph reveals:

The minimum threshold for a collaboration edge is two shared productions. Single co-credits are noise; repeated collaborations are signal.

Every contact comes with context: shared history with the agency, collaboration patterns from the graph, recent activity, and relationship trajectory. When you open a conversation with a casting director, you see what they’ve been working on, which of your actors they’ve booked before, and what their network looks like right now.

Actor profiles are enriched from multiple sources — Sanity CMS for bios and headshots, IMDB/TMDB for film and TV credits, and Scendatabasen for Swedish theater work. 321 film/TV credits and 370 theater credits are tracked, with semantic search across the full roster using three specialized vector spaces (biography, acting style, and skills).

The system watches for signals that indicate when to reach out: new productions being announced, incoming casting emails, and shifts in the industry landscape. When a signal appears, it’s automatically enriched with graph context — who is involved, how they connect to the agency’s network, and which relationships are relevant.

Signal sources:

When the system surfaces an opportunity — a relevant signal, a warming relationship, an incoming brief — it helps close the loop. For casting emails, it extracts role descriptions, matches them against the actor roster using semantic search, and generates personalized draft replies. For proactive outreach, it tracks campaigns, recipient engagement, and follow-up timing. Everything flows back into the relationship context, so the next interaction is even more informed.

You’re about to call a casting director. Instead of going in cold or relying on memory, you check the bot: their recent productions, who they collaborate with, which of your actors they’ve booked before, what their network looks like right now. The call becomes a conversation between people who know each other’s world, not a pitch from a stranger.

An email arrives with a casting brief: “Woman, 35-45, vulnerable but strong, Swedish with possible dialect.” The system detects it, runs semantic search against the roster, and generates a draft reply with matched actors and personalized justifications. What used to take an hour of browsing profiles and writing emails is ready in minutes.

A quick /intel shows which relationships are warming — one casting director has been on 3 productions recently and has history with your actors. An industry signal shows activity in their network. Time for a friendly check-in, not because a brief arrived, but because you’re genuinely present in the relationship.

You meet a producer and recognize the name but can’t place the context. A quick /network on your phone shows their production history, key collaborators, and which casting directors they work with. Now you can have a real conversation — not “so what do you do?” but “I saw you worked with [director] on [production], how was that?”

PostgreSQL can answer “who worked on this production?” Neo4j can answer “who is most likely to work with this director next?” Relationship patterns — production bubbles, strengthening collaborations, network clusters — are invisible in rows and columns but obvious in a graph. The dual-database design gives the best of both worlds.

The initial instinct was to optimize for speed — respond to briefs faster than anyone else. But the real competitive advantage is being the agency that doesn’t need the brief to start the conversation. When you’re already in the casting director’s mind because you’ve been relevant and present, the brief is a formality. The graph makes this kind of sustained presence possible across dozens of relationships.

The “production bubble” insight — that Swedish directors repeatedly work with the same collaborators — is worth more than any amount of ML refinement. A simple weighted Cypher query based on this insight outperforms generic recommendation algorithms because it encodes how the industry actually works.

Casting decisions happen between meetings, on the phone, on set. A web dashboard that requires opening a laptop is a CRM that sits unused. Telegram is already in your pocket. Every command is a live query — no loading screens, no navigation, no login. The interface disappears and you’re just having a conversation with your network intelligence.

The pipelines handle the tedious work: syncing credits, refreshing the graph, monitoring inboxes, tracking engagement. But the outreach itself stays human. The system tells you who to call and gives you context for the conversation. It doesn’t make the call for you. Automation serves the relationship, not the other way around.