datallm

A DataLLM allows a user to interact with a database using prompts and getting user-friendly answers. It works as follows -

Convert metadata from a DB into embeddings using an Embeddings API (metadata_embedder.py)
Store the embeddings in a document retriever (metadata_embedder.py)
When the user enters a prompt, convert the prompt into an embedding (similarity_search.py)
Compare the uses prompt embedding with the DB metadata embeddings using a cosine similarity search and return the most relevant tables for answering the prompt (similarity_search.py)
Send the user prompt and the table metadata to an LLM and ask to generate a SQL query (sql_generator.py)
Execute the SQL query in the DB and fetch the results (query_executor.py)
Send the user prompt and the SQL query results to an LLM and ask to generate a user-friendly response (generate_response.py)
Return the user-friendly response to the user (generate_response.py) --END--

Name		Name	Last commit message	Last commit date
Latest commit History 8 Commits
.venv		.venv
__pycache__		__pycache__
datallm2-frontend		datallm2-frontend
.DS_Store		.DS_Store
README.md		README.md
app.py		app.py
clear_tables.py		clear_tables.py
data.xlsx		data.xlsx
data_generator.py		data_generator.py
data_loader.py		data_loader.py
data_old.xlsx		data_old.xlsx
database.db		database.db
datallm2.code-workspace		datallm2.code-workspace
db_manager.py		db_manager.py
drop_tables.py		drop_tables.py
embeddings.json		embeddings.json
generate_response.py		generate_response.py
metadata_embedder.py		metadata_embedder.py
preview.webp		preview.webp
query_executor.py		query_executor.py
setup_database.py		setup_database.py
similarity_search.py		similarity_search.py
sql_generator.py		sql_generator.py
test_imports.py		test_imports.py
~$data.xlsx		~$data.xlsx

Provide feedback