A very rough demo showing how you can use an LLM to ask questions of a PDF document of your choice.
You will need to add the following to the .env file:
OPENAI_API_KEY=yourApiKey
PINECONE_API_KEY=yourApiKey
PINECONE_ENVIRONMENT=yourEnvName e.g. gcp-starter
NOTE: you can use alternatives to OpenAI and Pinecone
On the first run you may want to keep the main function main.js (line 208) looking like this so you run through all the steps:
(async () => {
await createPineconeIndex();
const loadedDocs = await loadDocuments();
const chunks = await createChunks(loadedDocs);
const embeddings = await createEmbeddings(chunks);
await updatePineconeIndex(chunks, embeddings);
// console.log("Uncomment this next line to delete the index");
// await deleteThisIndex();
await queryPineconeThenGPT3();
})();
But after a successful first run which creates the index, chunks and embeddings, it is probably best to comment out all but the queryPineconeThenGPT3() function.
The following lines in main.js (lines 16-20) also need editing to your particular set-up:
const question = "When do I have to submit a Form R?";
const indexName = "gold-guide-9"; // I named this index via the Pinecone UI
const gg9EmbeddingsLocation = "embeddings/gg9.json";
node main will run the main function.
The result should be a sensible answer to your question in the console!
e.g.
Querying Pinecone vector store...
Pinecone results to the question: {
matches: [
{
id: 'batch300-chunk5',
score: 0.82536006,
values: [Array],
sparseValues: undefined,
metadata: [Object]
},
{
id: 'batch300-chunk4',
score: 0.816575,
values: [Array],
sparseValues: undefined,
metadata: [Object]
},
{
id: 'batch100-chunk2',
score: 0.810189,
values: [Array],
sparseValues: undefined,
metadata: [Object]
},
{
id: 'batch100-chunk1',
score: 0.809418201,
values: [Array],
sparseValues: undefined,
metadata: [Object]
},
{
id: 'batch300-chunk3',
score: 0.806505084,
values: [Array],
sparseValues: undefined,
metadata: [Object]
}
],
namespace: ''
}
Pinecone returned 5 matches.
Top result from Pinecone: {
id: 'batch300-chunk5',
score: 0.82536006,
values: [
-0.00627459818, -0.000171131396, 0.0190629605, -0.0201321747,
-0.0170370806, 0.0108328266, -0.0122678243, -0.00528276153,
-0.0217641331, -0.0186409019, 0.00977768097, 0.0128305685,
-0.0122256186, -0.00909535307, -0.000311927375, -0.00529683,
0.016769778, 0.00581736863, -0.0023441813, 0.00901094172,
-0.0157990437, -0.00549730752, -0.0414742492, -0.00964402873,
-0.00238462863, 0.0207511932, 0.0125280935, -0.0170511492,
0.00725236628, 0.00732974336, -0.000721016, -0.00974954385,
0.0113603994, -0.0155176716, -0.031260442, -0.0112619186,
-0.0268710367, 0.00247959164, 0.019175509, -0.00189926173,
0.0116628744, 0.00247431593, 0.00045283325, 0.017684238,
0.00746339513, -0.0122396871, -0.000488004764, -0.029741032,
-0.0221017785, 0.0478613973, 0.00853964314, -0.00960885733,
-0.0308383834, -0.0136254448, 0.0384635665, 0.018317325,
-0.01074138, 0.0243949629, -0.0000438544812, -0.00673182774,
0.00327798515, -0.0167416409, -0.0117261829, 0.018542422,
-0.00741415517, 0.016769778, -0.0259706452, 0.0173043851,
-0.0119723836, 0.0186268333, 0.0184298735, 0.0126828477,
0.0227630045, 0.000423816731, 0.0155317402, -0.0311478935,
-0.0158412494, -0.0242120698, -0.00794172753, -0.000246860058,
-0.00756187551, -0.0176983066, -0.000417002273, 0.0181766376,
0.00458636554, -0.004952149, -0.00370707759, 0.0201181062,
-0.0268569682, -0.000733326073, -0.00614094641, 0.0545017794,
0.0112759871, 0.0000447612474, -0.0146735553, -0.00976361241,
-0.00508580077, 0.0321608335, -0.00653135031, -0.0188097265,
... 1436 more items
],
sparseValues: undefined,
metadata: {
loc: '{"lines":{"from":1,"to":1}}',
pageContent: '83 | Version: GG9 incorporating the Purple Guide, August 2022 called to a support meeting with their Postgraduate Dean/RO or their nominated deputy to discuss the reasons for non-submission and to clarify next steps if the situation is not rectified. 4.127 If a trainee submits or resubmits a completed Form R or SOAR within the two-week timeframe, they receive an ARCP outcome appropriate for their educational progression and alignment with the GMC’s standards in Good Medical Practice. 4.128 If the trainee still fails to submit a satisfactorily completed Form R or SOAR after two weeks and this is the first time that this situation has arisen in the training programme, for foundation, core, specialty and general practice trainees, an Outcome 2 (not applicable in foundation), 3 or 4 will be issued (according to training progression). A note is made on the trainee’s record that they did not submit a completed Form R or SOAR. An Outcome 1 or 6 is not awarded, even if there are no training',
source: 'Page: 83'
}
}
Top result's metadata from Pinecone: {
loc: '{"lines":{"from":1,"to":1}}',
pageContent: '83 | Version: GG9 incorporating the Purple Guide, August 2022 called to a support meeting with their Postgraduate Dean/RO or their nominated deputy to discuss the reasons for non-submission and to clarify next steps if the situation is not rectified. 4.127 If a trainee submits or resubmits a completed Form R or SOAR within the two-week timeframe, they receive an ARCP outcome appropriate for their educational progression and alignment with the GMC’s standards in Good Medical Practice. 4.128 If the trainee still fails to submit a satisfactorily completed Form R or SOAR after two weeks and this is the first time that this situation has arisen in the training programme, for foundation, core, specialty and general practice trainees, an Outcome 2 (not applicable in foundation), 3 or 4 will be issued (according to training progression). A note is made on the trainee’s record that they did not submit a completed Form R or SOAR. An Outcome 1 or 6 is not awarded, even if there are no training',
source: 'Page: 83'
}
Now querying GPT-3 with question... : When do I have to submit a Form R?
Answer: You are required to submit a Form R annually, except for F1 doctors who do not need to submit it. The completed Form R needs to be submitted before your ARCP (Annual Review of Competence Progression) to ensure your registration is renewed on an annual basis.