Invoke PostgreSQL function with vector parameter using Prisma $queryRaw

invoke_vector_function_from_prisma.sql

--I'm assuming you have already added the 'pgvector' extension to PostgreSQL (https://github.com/pgvector/pgvector)

--looks wierd but you invoke the function using select, representing the vector as a string literal..
select * from public.test_function('[1.03, 2.45, 3.56]', .06);

my_service.ts

import { Prisma} from '@prisma/client';
import prisma_client from 'prisma.client';

//for stateless 'DAO' style services like this, I have started using Namespaces..
export namespace MyService {
  //use $queryRaw to build up the sql, replicate the wierd string literal array.  Prisma.raw is used to tell prisma not to mess with the literal
  export async function callTestFunction(
    query_embedding: number[],
    similarity_threshold: number = 0.05
  ) {
    return prisma_client.$queryRaw`
    SELECT * FROM public.test_function(
      ${Prisma.raw(`'[${query_embedding.join(',')}]'`)}, 
      ${similarity_threshold}::double precision
    )`;
  }
}

prisma.client.ts

import { PrismaClient } from '@prisma/client';

const prisma_client = new PrismaClient();
export default prisma_client;

test_function.sql

CREATE OR REPLACE FUNCTION public.test_function(
  query_embedding vector,
  similarity_threshold double precision DEFAULT 0.05 
)
 RETURNS TABLE(test text)
 LANGUAGE sql
 STABLE
AS $function$
  --you would normally use query_embedding to do a distance query using <=> etc... but for test just return rubbish
  select 'passed' as test;
$function$