Manamesa

Problem

Magic players rely on fragmented tools for deckbuilding, card search, price lookup, collection tracking, rules context, and strategy. The work is repetitive, spread across many interfaces, and hard to connect to a player's actual decks and collection.

• Deckbuilding often happens separately from card data, prices, and collection ownership.
• Card search still requires players to translate intent into filters and keywords.
• Price and trade decisions are difficult to evaluate quickly, especially in local markets.
• Strategic recommendations rarely understand the player's actual deck, budget, and inventory.

Solution

Manamesa is one conversational AI assistant connected to deterministic engines for decks, cards, prices, collections, audio, and OCR. The player describes what they need; the system interprets intent, calls structured tools, updates state, and responds with useful next steps.

• Manage decks through natural language.
• Search cards semantically by gameplay intent.
• Consult price intelligence, initially focused on Brazil.
• Use memory and collection context to improve recommendations over time.

An AI project

Manamesa is an AI project built around agent orchestration, structured tool calls, semantic search, audio transcription, image/OCR workflows, recommendation systems, and price intelligence. The LLM interprets player intent; backend tools execute critical actions safely.

• Agent layer for intent interpretation and multi-step task planning.
• Structured tool calls for deck, card, price, and collection actions.
• Semantic search for card discovery and strategic recommendations.
• Audio and future image/OCR pipelines for multimodal input.

Project phases

The roadmap starts with a narrow usable core and expands toward a full AI platform for deck management, analysis, collection intelligence, and deck building.

1. Phase 1: user context, create deck, add/remove cards, view deck.
2. Phase 2: card search, price lookup, audio processing.
3. Phase 3: import/export and deck sharing.
4. Phase 4: AI deck analysis.
5. Phase 5: personal collection.
6. Phase 6: card scanner/OCR.
7. Phase 7: AI deck builder.

Gateway-agnostic entry. Backend as the source of truth.

Gateways should remain replaceable. The durable product is the backend: user context, orchestration, tools, databases, caches, queues, logs, and decisions that make the assistant reliable across channels.

• Backend as the source of truth.
• LLM as an interpretation layer, not a rules database.
• Deterministic tools for critical actions.
• Observability for messages, errors, and agent decisions.

Program fit

The project aligns with AI, cloud, and innovation programs because it combines an AI-first product direction with agent orchestration, LLM workflows, scalable asynchronous jobs, data/search/cache architecture, observability, responsible operations, and future multimodal OCR/image processing.