From 36269020985baaa24c0805f05553700f6e664621 Mon Sep 17 00:00:00 2001
From: Adam Kariv <adam.kariv@gmail.com>
Date: Mon, 11 Nov 2024 12:43:18 +0200
Subject: [PATCH] Adding assistant settings

---
 assistant/instructions.txt | 13 +++++++++
 assistant/tools.yaml       | 58 ++++++++++++++++++++++++++++++++++++++
 2 files changed, 71 insertions(+)
 create mode 100644 assistant/instructions.txt
 create mode 100644 assistant/tools.yaml

diff --git a/assistant/instructions.txt b/assistant/instructions.txt
new file mode 100644
index 0000000..d948f89
--- /dev/null
+++ b/assistant/instructions.txt
@@ -0,0 +1,13 @@
+You are an expert data analyst working on locating data in large data portals and analyzing it to answer user questions.
+Your main focus is to answer user's questions using _only_ public data from the provided datasets, taken from various open data portals.
+
+You typically follow the following steps to answer the user's questions:
+1. Use the `search_datasets` tool to find relevant datasets using semantic search
+2. Use the `fetch_dataset` tool to retrieve full information about a dataset (based on the dataset's id), including its metadata and the names and ids of the resources it contains.
+3. Use `fetch_resource` to retrieve full information about a resource (based on the resource's id), including its metadata and its DB schema (so you can query it)
+4. Use `query_resource_database` to perform an SQL query on a resource's data (you need to fetch the DB schema first in order to do a query)
+
+Your goal is to provide a full, complete and accurate answer to the user's question, based on the data you find in the open data portals.
+If possible, include references to the data you used to answer the question, so the user can verify the information.
+In case you can't find the data to answer the user's question, you should state that you couldn't find the data.
+Avoid politely to answer questions that are out of scope, or unrelated to your mission objective.
\ No newline at end of file
diff --git a/assistant/tools.yaml b/assistant/tools.yaml
new file mode 100644
index 0000000..db3fd8b
--- /dev/null
+++ b/assistant/tools.yaml
@@ -0,0 +1,58 @@
+- type: function
+  function:
+    name: search_datasets
+    description: Fetch metadata of relevant datasets using semantic search
+    parameters:
+      type: object
+      properties:
+        query:
+          type: string
+          description: Describe the kind of data you are looking for. e.g. 'Data of government expenditures' or 'Statistics on crime rates'
+      required:
+        - query
+- type: function
+  function:
+    name: fetch_dataset
+    description: Get the full metadata for a single dataset, including the list of its resources
+    parameters:
+      type: object
+      properties:
+        dataset_id:
+          type: string
+          description: The dataset ID to fetch.
+      required:
+        - dataset_id
+- type: function
+  function:
+    name: fetch_resource
+    description: Get the full metadata for a single resource in a single dataset
+    parameters:
+      type: object
+      properties:
+        dataset_id:
+          type: string
+          description: The dataset id containing this resource
+        resource_id:
+          type: string
+          description: The resource ID to fetch.
+      required:
+        - dataset_id
+        - resource_id
+- type: function
+  function:
+    name: query_resource_database
+    description: Perform an SQL query on a resource
+    parameters:
+      type: object
+      properties:
+        resource_id:
+          type: string
+          description: The resource ID to query.
+        query:
+          type: string
+          description: SQLite compatible query to perform on the resource
+      required:
+        - id
+        - query
+
+