Skip to content

Latest commit

 

History

History
142 lines (105 loc) · 3.89 KB

README.md

File metadata and controls

142 lines (105 loc) · 3.89 KB
Raw

Langroid API Client

Python client for Langroid REST API.

Installation

Work in a virtual environment, and use Python 3.11.

python3 -m venv .venv
. ./.venv/bin/activate
pip install langroid-client

Environment Setup

Have your OpenAI API Key ready (it should be gpt-4 capable).

Set these env vars in your .env file placed in the root of this repo.

OPENAI_API_KEY=your-api-key-with-no-quotes
INTELLILANG_BASE_URL="https://langroid-server-zb43tal5mq-uk.a.run.app"

Run the example script

Run this from the root of this repository as follows. The above environment vars will be automatically loaded from the .env file.

Look into this example script to see how to use the client (also see details below).

python3 examples/example_usage.py

Or if you don't have an .env file, you can pass the vars directly at the cmd line:

INTELLILANG_BASE_URL="https://langroid-server-zb43tal5mq-uk.a.run.app" \
    OPENAI_API_KEY=your-api-key-with-no-quotes \    
    python3 examples/example_usage.py

Usage of the Python API client

Before using these, first set up the Client object, using the INTELLILANG_BASE_URL env var.

Note, the API handles docs in a variety of formats-- pdf, doc, docx, txt.

from langroid_client import LangroidClient
client = LangroidClient(INTELLILANG_BASE_URL)

Extract requirements

success, reqs_jsonl = client.intellilang_extract_reqs(
    reqs_path="path/to/requirements.pdf", # the requirements document to extract from
    candidate_path="path/to/candidate.pdf", # example of a candidate/proposal file
    params={"num": 3}, # number of requirements to extract
    openai_api_key="your-api-key",
    doc_type="rfp", # or "resume"
)
    """
    Extract requirements from a specification document.

    Args:
        reqs_path (str): Path to the requirements document.
        candidate_path (str): Path to the candidate document.
        params (Dict[str, Any]): Extraction parameters.
        openai_api_key (str): OpenAI API key.
        doc_type (str): Type of document (rfp or resume).

    Returns:
        Tuple[bool, bytes|str]:
            A tuple containing a boolean indicating success and the extracted
            requirements in jsonl format
    """

The returned success is a bool flag indicating success or not.

Save requirements to a jsonl format file

extracted_reqs_jsonl = "/tmp/out.jsonl"
with open(reqs_jsonl, "wb") as output_file:
    output_file.write(reqs_jsonl)

Evaluate candidate docs w.r.t. extracted reqs

import json
success, (scores, evals) = client.intellilang_eval(
    extracted_reqs_jsonl, # above extracted requirements jsonl file
    ["/path/to/candidate1.pdf", "/path/to/candidate2.pdf"], # list of candidate files
    params=dict(start_idx=1, cost=30.0), # leave these as a default
    openai_api_key="openai-api-key",
    doc_type="rfp",
)
"""
Evaluate candidates based on extracted requirements.

Args:
    reqs_path (str): Path to the extracted requirements file (jsonl file)
    candidate_paths (List[str]): Paths to the candidate documents.
    params (Dict[str, Any]): Evaluation parameters.
    openai_api_key (str): OpenAI API key.
    doc_type (str): Type of document (rfp or resume).

Returns:
Tuple[bool, Tuple[List[Dict[str, Any]], List[Dict[str, Any]] | str]:
            A tuple containing a boolean indicating success and tuple
of lists of scores and evaluations
"""

Besides the success flag, a tuple of two list-of-dicts is returned:

  • scores represents the synopsis table of scores of all candidates
  • evals represents a table containing detailed evals of all candidates (best to look at the example output to understand the structure)
print(f"success: {success}")
# print scores
print("Scores:")
for score in scores:
    print(json.dumps(score))

# print evals
print("Evaluations:")
for eval in evals:
    print(json.dumps(eval))