Short Project Descriptions + Fetch Projects By Category

web/components/ProjectModal.tsx

@@ -118,6 +118,11 @@ const ProjectModal = ({
       contentStyles={{ padding: "p-4 sm:p-6 pt-0" }}>
       {strategy && (
         <div className='space-y-4'>
+            <div className='bg-indigo-50 p-3 rounded-xl space-y-3'>
+              <div className={clsx("text-sm leading-normal")}>
+                {strategy.project.short_description}
+              </div>
+            </div>
           <div className='bg-indigo-50 p-3 rounded-xl space-y-3'>
             <div
               className='flex items-center justify-between pb-2 border-b border-indigo-600 group'>

web/supabase/migrations/20240219173654_get_categories.sql

@@ -0,0 +1,2 @@
+CREATE VIEW unique_categories_views AS SELECT DISTINCT unnest(categories) AS category
+FROM "public"."projects";

web/utils/ethereum/supportedErc20TokensByNetwork.ts

@@ -126,5 +126,17 @@ export const supportedErc20TokensByNetwork: SupportedTokensInformation = {
       decimals: 6,
       name: "USDT"
     }
+  },
+  Base: {
+    USDC: {
+      address: "0x833589fcd6edb6e08f4c7c32d4f71b54bda02913",
+      decimals: 6,
+      name: "USDC",
+    },
+    DAI: {
+      address: "0x50c5725949a6f0c72e6c4a641f24049a917db0cb",
+      decimals: 18,
+      name: "DAI",
+    }
   }
 };

workers/fund_public_goods/db/tables/projects.py

@@ -1,4 +1,4 @@
-from typing import Any, Dict
+from typing import Any
 from fund_public_goods.lib.strategy.models.answer import Answer
 from supabase import PostgrestAPIResponse
 from fund_public_goods.db.entities import Projects
@@ -66,39 +66,9 @@ def get(
         logo=data["logo"]
     )
 
-def get_projects(range_from: int, range_to: int) -> PostgrestAPIResponse[Dict[str, Any]]:
-    db = create_admin()
-    return (
-        db.table("projects")
-        .select(
-            "id, updated_at, title, description, website, keywords, categories, short_description, twitter, logo, applications(id, recipient, round, answers)"
-        )
-        .range(range_from, range_to)
-        .execute()
-    )
-
-def get_all_projects() -> list[dict[str, Any]]:
-    all_results: list[dict[str, Any]] = []
-    current_from = 0
-    page_size = 999
-    while True:
-        current_to = current_from + page_size
-        results = get_projects(current_from, current_to).data
-        all_results.extend(results)
-
-        if len(results) < page_size:
-            break
-
-        current_from += page_size
-
-    return all_results
-
-def fetch_projects_data() -> list[tuple[Projects, list[Answer]]]:
-    data = get_all_projects()
-
+def sanitize_projects_information(projects: list[dict[str, Any]]) -> list[tuple[Projects, list[Answer]]]:
     projects_with_answers: list[tuple[Projects, list[Answer]]] = []
-
-    for item in data:
+    for item in projects:
         answers: list[Answer] = []
 
         for application in item.get("applications", []):
@@ -127,3 +97,37 @@ def fetch_projects_data() -> list[tuple[Projects, list[Answer]]]:
         projects_with_answers.append((project, answers))
 
     return projects_with_answers
+
+
+def get_unique_categories() -> list[str]:
+    db = create_admin()
+    response: PostgrestAPIResponse[list[dict[str, str]]] = (
+        db.table("unique_categories_views").select("*").execute()
+    )
+    if not response.data:
+        return []
+
+    categories = []
+
+    for row in response.data:
+        categories.append(row["category"]) # type: ignore
+
+    return categories
+
+def fetch_projects_by_category(categories: list[str]) -> list[tuple[Projects, list[Answer]]]:
+    results = get_projects_from_description(categories).data
+    sanitized_projects = sanitize_projects_information(results) 
+    return sanitized_projects
+
+def get_projects_from_description(categories: list[str]):
+    db = create_admin()
+    request = (
+        db.table("projects")
+        .select(
+            "id, updated_at, title, description, website, keywords, categories, short_description, twitter, logo, applications(id, recipient, round, answers)"
+        )
+        .ov("categories", categories)
+        .execute()
+    )
+
+    return request

workers/fund_public_goods/lib/strategy/utils/categorize_prompt.py

@@ -9,17 +9,17 @@
 
 {categories}
 
-A user's prompt can match to more than one category. Be strict with assigning categories.
-Return a max of {n} categories, you can reutrn less if the project really only matches less than {n} categories.
-
 Respond strictly with a comma-separated list of categories, without quotes. Do not change the wording
 or casing of the categories, return them exactly as they are written in the list above.
 
+You must make sure that the categorization of the prompt is extensive enough so projects can be retrieved
+based on these categories.
+
 Prompt: {prompt}
 """
 
 
-def categorize_prompt(prompt: str, n: int) -> list[str]:
+def categorize_prompt(prompt: str, categories: list[str]) -> list[str]:
     categorize_prompt = ChatPromptTemplate.from_messages([
         ("system", categorize_prompt_template),
     ])
@@ -29,8 +29,7 @@ def categorize_prompt(prompt: str, n: int) -> list[str]:
 
     categories = [c.strip() for c in categorize_chain.invoke({
         "prompt": prompt,
-        'n': n,
-        "categories": "\n".join(f"- {category}" for category in PROJECT_CATEGORIES)
+        "categories": "\n".join(f"- {category}" for category in categories)
     }).split(',')]
 
     return categories

workers/fund_public_goods/lib/strategy/utils/fetch_matching_projects.py

@@ -1,13 +1,15 @@
 from fund_public_goods.db.entities import Projects
-from fund_public_goods.db.tables.projects import fetch_projects_data
+from fund_public_goods.db.tables.projects import fetch_projects_by_category, get_unique_categories
 from fund_public_goods.lib.strategy.models.answer import Answer
+from fund_public_goods.lib.strategy.utils.categorize_prompt import categorize_prompt
 from fund_public_goods.lib.strategy.utils.get_top_matching_projects import get_top_matching_projects
 from fund_public_goods.lib.strategy.utils.utils import get_latest_project_per_website
-from fund_public_goods.workflows.egress_gitcoin.upsert import sanitize_url
 
 
 def fetch_matching_projects(prompt: str) -> list[tuple[Projects, list[Answer]]]:
-    fetched_projects = fetch_projects_data()
+    prompt_categories = categorize_prompt(prompt, get_unique_categories())
+    fetched_projects = fetch_projects_by_category(prompt_categories)
+
     answers_by_id = { project.id: answers for (project, answers) in fetched_projects }
     projects = [project for (project, _) in fetched_projects]
 

workers/fund_public_goods/lib/strategy/utils/get_top_matching_projects.py

@@ -48,7 +48,7 @@ def rerank_top_projects(prompt: str, projects: list[Projects]) -> list[Projects]
     reranking_chain = reranking_prompt | llm | StrOutputParser()
 
     separator = "\n-----\n"
-    
+
     top_ids_res = reranking_chain.invoke({
         "prompt": prompt,
         "separator": separator,
@@ -132,11 +132,7 @@ def filter_projects_by_categories(projects: list[Projects], categories: list[str
 
 def get_top_matching_projects(prompt: str, projects: list[Projects]) -> list[Projects]:
     projects_by_id = {project.id: project for project in projects}
-
-    prompt_categories = categorize_prompt(prompt, 2)
-    projects_with_categories = filter_projects_by_categories(projects, prompt_categories)
-
-    all_projects_collection = create_embeddings_collection(projects_with_categories)
+    all_projects_collection = create_embeddings_collection(projects)
     matches = all_projects_collection.similarity_search(prompt, k=300)
     matched_project_ids = [match.metadata["id"] for match in matches]
     unique_ids = get_top_n_unique_ids({prompt: matched_project_ids}, 30)