By Heli and Marko Helskyaho

WishList Resolver: The Select AI Agent that turns vague dreams into deliverable gifts.

The WishList Resolver is the magical middleman between a child’s wild imagination and Santa’s finite toy inventory. It’s a fully autonomous agent that takes fuzzy, last-minute wishes (like “I want a puppy that never grows” or “a robot that sings Jingle Bells”) and translates them into precise, stock-available gift recommendations, all while respecting the Korvatunturi’s strict data privacy rules.

This agent uses natural language understanding, hybrid search, and tool orchestration to handle ambiguous queries in real-time, making it a blueprint for e-commerce recommendation engines or customer intent resolvers in any enterprise.

How does this agent work?

  1. Parses the wish: 
    • Starts with natural language input from a child (or parent), using an LLM to break it down into key attributes (age-appropriate, category, features, budget constraints, e.g., “under $50 for a 7-year-old”).
  2. Searches the catalog: Leverages Oracle’s unified hybrid search to query:
    • Relational tables for toy specs (price, age range, availability).
    • Vector embeddings of product descriptions to match semantic intent (for example, “eternal puppy” → stuffed animals with “low-maintenance” tags).
    • JSON docs for user history (past gifts, allergies) and graph analytics for trending items.
  1. Reasons and ranks: Applies the ReAct pattern to iterate:
    • Scores matches based on similarity and stock levels.
    • Checks for substitutes if exact matches are out (for example, cannot find real flying carpet, suggest a drone with lights or a carper with magical pattern).
    • Factors in logistics like “deliverable by Christmas Eve” via integrated route data.
  2. Resolves and responds: 
    • Outputs a curated list with reasons, images (if multimodal), and confirmation updating the manifest via PL/SQL if approved. Everything stays in the database for zero-leak security.

EXEC DBMS_CLOUD_AI_AGENT.set_team(team_name  => ‘SantasAgents’);

SELECT AI AGENT I want a magical flying carpet for my dollhouse, but it has to be sparkly and safe for a 6-year-old girl in London

Sample response:

Ho ho ho! Let’s make that wish come true, little dreamer! Your “magical flying carpet” sounds like a cozy adventure for your dollhouse. Scanning the Korvatunturi catalog…

Top Match (92% Fit): SparkleWeave Fairy Rug – Soft fabric glider with LED lights and safe, detachable wings. Age 5+, £24.99. In stock: 1,247 units (London warehouse ETA: 24-Dec 23:45). Why? Matches “sparkly” (glitter threads) and “flying” (gentle ramp launch—no batteries needed). Backup Option (78% Fit): Enchanted Cloud Blanket – Plush throw with hidden pockets for doll “flights.” Age 4+, £18.50. Stock: 892 units. Safer for tiny hands, but less “carpet-y.

Pro Tip: Pairs perfectly with your 2024 WishList fave, the Mini Doll Jetpack! Shall I add to the sleigh manifest?

Merry matching! 🎁✨ (Sources: TOY_CATALOG_VECTORS, CHILD_WISH_HISTORY)

Another example:

l_response := DBMS_CLOUD_AI_AGENT.RUN_TEAM(
team_name => ‘SantasAgents’,
user_prompt => ‘Reroute for Tokyo: 5 new kids added in Shibuya district due to family move. Factor in 20mph winds from the north.’
,params => ‘{“conversation_id”:”‘ || l_conversation_id || ‘”}”
);

Sample Response:

Route recalculation underway, Santa! Analyzing 847,392 Asia-Pacific stops…

Original ETA Impact: +14 minutes delay from Tokyo insert (Shibuya cluster: 5 new houses, avg. gift weight 2.3kg). Winds at 20mph NNE could add 8% drag on Sleigh-3.

Optimized Plan:  

Shift Tokyo window to 23:47-00:12 JST (post-wind peak).  

Reroute via Yokohama detour: Total distance -2.7km, fuel savings 1.4 carrots equivalent.  

Reindeer swap: Dasher leads (high-wind specialist); Prancer rests.  

New sequence: House #JP-456 → #JP-789 (new Shibuya) → #JP-012 (group +12% efficiency).

Global impact: Zero net delay. Full manifest updated in SLEIGH_PLAN table.

Clear skies ahead—ho ho hover! 🛷🌪️ (Sources: WORLD_GRAPH, WEATHER_REST, REINDEER_VITALS)

Why It’s Technically Impressive (and Elf-Approved)

  • The team notices this is a wish and gives the task to agent-elf WishList Resolver
  • Handles ambiguity like a pro: Uses vector search for semantic matching (for example, “flying carpet” → airborne toys) and NL2SQL for precise filters (for example, “age <=6 AND sparkly=1”).
  • Secure and scalable: Inherits row-level security, only verified kids see personalised results.
  • Audit-ready: Logs every resolution for “Why did Timmy get a drone instead?” queries from skeptical parents or regulators.
  • Real-world power: Beyond holidays, it’s like an AI product recommender for Amazon, grounded in your inventory data, no external APIs needed unless you add them.

It’ll resolve your queries faster than you can say “Jingle all the way!” 🎄 🎁