MIME-Version: 1.0
From: Allan Kamau <kamauallan@gmail.com>
Date: Tue, 10 Dec 2024 15:59:00 +0300
Message-ID: <CAF3N6oQ4GeF93L34FJ=E5GoYBHXv7H4H0gK4FANgv=fHORoN4Q@mail.gmail.com>
Subject: Formulating SQL queries via Retrieval Augmented Generation (RAG).
To: Postgres General Postgres General <pgsql-general@postgresql.org>
Content-Type: multipart/alternative; boundary="0000000000007405030628ea0c53"
Archived-At: <https://www.postgresql.org/message-id/CAF3N6oQ4GeF93L34FJ%3DE5GoYBHXv7H4H0gK4FANgv%3DfHORoN4Q%40mail.gmail.com>
Precedence: bulk

--0000000000007405030628ea0c53
Content-Type: text/plain; charset="UTF-8"

I am not sure if this is the right place to post a PostgreSQL + Retrieval
Augmented Generation (RAG) question, but I have tried other sites but I did
not receive any response.

I have managed to use RAG to query data in a PostgreSQL table using Python
and the lanchain_community module.

But I have been struggling for weeks trying to find a solution to using RAG
to perform WITH RECURSIVE CTE on two tables (modeling node and edge graph
data).
The SQL queries generated by RAG for this graph data model are error prone.
Is there an example somewhere where RAG has been used to successfully
generate WITH RECURSIVE CTE on two table graph data?

Below is the main code section I am using to perform RAG on PostgreSQL.

    llm = ChatGroq(model="mixtral-8x7b-32768", temperature=0);

    chain = (
        RunnablePassthrough.assign(query=sql_chain).assign(
            schema=lambda _: db.get_table_info(),
            response=lambda vars: db.run(vars["query"]),
        )
        | prompt
        | llm
    )
    response_obj: langchain_core.messages.ai.AIMessage=chain.invoke({
        "question": user_query_str,
        "chat_history": chat_history_list,
    });


Any pointers are welcome.

-Allan

--0000000000007405030628ea0c53
Content-Type: text/html; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

<div dir=3D"ltr"><div dir=3D"ltr"><div dir=3D"ltr"><div>I am not sure if th=
is is the right=C2=A0place to post a PostgreSQL + Retrieval Augmented Gener=
ation (RAG) question, but I have tried other=C2=A0sites but I did not recei=
ve any response.</div><div><br></div>I have managed to use RAG to query dat=
a in a PostgreSQL table using Python and the lanchain_community module.<div=
><br><div>But I have been struggling for weeks trying to find a solution to=
 using RAG to perform WITH RECURSIVE CTE on two tables (modeling node and e=
dge graph data).</div><div>The SQL queries generated by RAG for this graph =
data model are error prone.</div><div>Is there an example somewhere=C2=A0wh=
ere RAG has been used to successfully generate WITH RECURSIVE CTE on two ta=
ble graph data?</div><div><div><br>Below is the main code section I am usin=
g to perform RAG on PostgreSQL.</div><div><div><br></div><div><pre style=3D=
"margin-top:0px;margin-bottom:0px;border:0px;font-variant-numeric:inherit;f=
ont-variant-east-asian:inherit;font-stretch:inherit;vertical-align:baseline=
;box-sizing:inherit;width:auto;max-height:600px;overflow:auto;color:rgb(12,=
13,14)"><code style=3D"margin:0px;padding:0px;border:0px;font-style:inherit=
;font-variant:inherit;font-weight:inherit;font-stretch:inherit;line-height:=
inherit;vertical-align:baseline;box-sizing:inherit;background-color:transpa=
rent;white-space:inherit;border-radius:0px">    llm =3D ChatGroq(model=3D&q=
uot;mixtral-8x7b-32768&quot;, temperature=3D0);
   =20
    chain =3D (
        RunnablePassthrough.assign(query=3Dsql_chain).assign(
            schema=3Dlambda _: db.get_table_info(),
            response=3Dlambda vars: db.run(vars[&quot;query&quot;]),
        )
        | prompt
        | llm
    )
    response_obj: langchain_core.messages.ai.AIMessage=3Dchain.invoke({
        &quot;question&quot;: user_query_str,
        &quot;chat_history&quot;: chat_history_list,
    });</code></pre></div><div><br></div><div><br>Any pointers are welcome.=
</div><div><br></div></div></div><div>-Allan</div></div></div>
</div>
</div>

--0000000000007405030628ea0c53--