clojure-datomic-adb-23ai-vector-search

What this demo is

This microservice receives a question about food and returns the 5 most-likely foods from the datomic db (based on similarity search from Oracle Autonomous 23ai).

Example question: "Gostaria de sugestões de comidas baianas"

Anwser:

1. Um(a) delicioso(a) moqueca, do litoral baiano, preparado artesanalmente e feito(a) com azeite de dendê, que aquece a alma
2. Um(a) saboroso(a) vatapá, do recôncavo baiano, cozido(a) lentamente e feito(a) com coco ralado, que derrete na boca
3. Um(a) crocante acarajé, típico da Bahia, frito(a) no azeite de dendê e recheado(a) com vatapá.
4. Um(a) tradicional caruru, feito(a) com quiabo e camarão, temperado(a) com azeite de dendê.
5. Um(a) cremoso(a) bobó de camarão, preparado(a) com mandioca e azeite de dendê, típico da culinária baiana.

This is a small, end‑to‑end demo of a Clojure microservice that combines:

• Clojure + Pedestal (HTTP)
• Datomic using SQL storage on Oracle
• Oracle Autonomous Database Vector Search (LangChain4j OracleEmbeddingStore)

Endpoints (open in your browser):

• /greet?name=World → simple greeting (check service is alive)
• /food?text=your+query → JSON with top‑5 food matches {id,name,description,score}

What you’ll see

• How an HTTP handler calls a vector search on Oracle, then hydrates results from Datomic
• Clean separation of pure functions and side‑effects, and explicit dependency injection

Quick start

1. Start the service

lein run

2. Open in your browser

• Greeting: http://localhost:8080/greet?name=World (check service is alive)
• Food search (Portuguese): http://localhost:8080/food?text=comida baiana com dende"
• Food search (English): http://localhost:8080/food?text="Brazilian seafood with dende"

3. Microservice output

• Greeting: a plain text “Hello World!” (service is alive)
• Food search (example)

{
  "query": "comida baiana com dende",
  "results": [
    { "id": 17, "name": "moqueca", "description": "Um(a) delicioso(a) moqueca, do litoral baiano, preparado artesanalmente e feito(a) com azeite de dendê, que aquece a alma.", "score": 0.86 },
    { "id": 203, "name": "vatapá",  "description": "Um(a) saboroso(a) vatapá, do recôncavo baiano, cozido(a) lentamente e feito(a) com coco ralado, que derrete na boca.", "score": 0.81 },
    { "id": 42, "name": "acarajé", "description": "Um(a) crocante acarajé, típico da Bahia, frito(a) no azeite de dendê e recheado(a) com vatapá.", "score": 0.78 },
    { "id": 58, "name": "caruru", "description": "Um(a) tradicional caruru, feito(a) com quiabo e camarão, temperado(a) com azeite de dendê.", "score": 0.75 },
    { "id": 99, "name": "bobó de camarão", "description": "Um(a) cremoso(a) bobó de camarão, preparado(a) com mandioca e azeite de dendê, típico da culinária baiana.", "score": 0.73 }
  ],
  "count": 5
}

How to initialize it

Prerequisites

You need to set up two databases before running this demo:

1. Oracle Autonomous Database 23ai with Vector Search enabled

   • Create an Oracle Autonomous Database 23ai instance (OCI cloud)
   • Ensure vector search is enabled (this feature comes pre-configured in 23ai)
   • Download wallet files from your ADB console
   • Oracle Instant Client + ADB Wallet (set TNS_ADMIN to the wallet directory)

2. Datomic Database (SQL storage on Oracle)

   • Install and configure Datomic Transactor
   • Configure Datomic to use Oracle as the SQL storage backend
   • Ensure Datomic can connect to your Oracle 23ai database
   • Note: The demo will automatically create the Datomic database schema

• Java 17+, Leiningen

Environment Setup

Set these environment variables (PowerShell example):

$env:TNS_SERVICE="<tns_service>"
$env:ORACLE_DB_USER="<oracle_db_user>"
$env:ORACLE_DB_PASSWORD="<oracle_db_pw>"
$env:TNS_ADMIN="<wallet_folder>"
$env:DATOMIC_DB_NAME="<datomic_db_name>"

Database Preparation

Before running the demo, you need to prepare both Datomic and Oracle databases:

Note on Data Generation: This demo uses an automated data generation function (make-food-dataset) that creates 200 Brazilian food items with rich descriptions. The generated data includes:

• Traditional Brazilian food names
• Detailed descriptions with cooking methods, ingredients, and cultural context
• Vector embeddings stored in Oracle for similarity search
• Metadata stored in Datomic for result hydration

If you want to use your own data instead, you can modify the dataset generation or replace the populate calls.

1. Install dependencies and start REPL:

lein deps
lein repl

2. Run the database setup script:

(require '[clojure-datomic-adb-23ai-vector-search.dbconf :as db])

;; Create configuration
(def cfg (db/make-config))
(def uris (db/build-uris cfg))

;; Ensure Datomic database exists
(db/ensure-datomic-db-exists! (:datomic-uri uris))

;; Connect to Datomic
(def conn (db/connect-datomic! (:datomic-uri uris)))

;; Create Oracle connection pool
(def pds (db/oracle-pool-data-source! {:jdbc-url (:jdbc-url uris)
                                       :db-username (:db-username cfg)
                                       :db-password (:db-password cfg)}))

;; Create Oracle embedding store (creates vector table if needed)
(def store (db/oracle-embedding-store! pds (:datomic-db-name cfg)))

;; Populate with demo data (200 Brazilian food items)
(db/populate-datomic-oracle-db! conn store (db/make-food-descriptions 200))

;; Verify setup
(println "Setup complete! Database contains" 
         (count (db/q '[:find (count ?e) :where [?e :food/name]] (db/db conn))) 
         "food items")

3. Exit REPL and start the service:

lein run

How it works (very short)

• /food handler embeds your query, searches Oracle Vector Store, gets top matches
• It looks up details in Datomic (stored on Oracle via SQL storage)
• Returns merged results as JSON

Troubleshooting

• TNS_ADMIN must point to a valid Oracle wallet directory
• Oracle Instant Client/JDBC must be installed
• Port 8080 must be free

References

• I followed tutorials and had help from several people, so thank you! Here I mention some:
• https://scicloj.github.io/clojure-data-tutorials/projects/ml/llm/vectorstore.html
• https://medium.com/oracledevs/retrieval-augmented-generation-rag-with-spring-ai-oracle-database-23ai-and-openai-61281b96d18a