ACRRU is a multi-functional LLM-Driven research & reporting framework that automates the data collection and summarization needed to score infrastructure entities and regions with respect to the CRMM.
For a given entity, ACRRU executes a custom research plan to gather the data needed to address CRMM capabilities.
ACRRU then evaluates the collected data sources and identifies all evidence of CRMM capability criteria.
The evidence is summarized by capability into a final report for the entity, with all data sources cited. Multiple provider summaries can be further summarized into a regional report.
Use the crmm_env.yml file to create a conda environment for the application.
As of version 0.2.0, local usage will require licenses or accounts to the following services:
- OpenAI
- Tavily
- Google Sheets API if you would like to set up ACRRU logging. NOTE: If you do not wish to use this logging, please set the
gsheet_logargument toFalse.
Please provide this information in a .env file at the root of the repository. Refer to env_example.txt for expected format.
As of version 0.2.0, only local runs that begin with research are supported.
Inputs are defined in text files within the inputs folder. Each line represents an entity and its geographic mapping, separated by a pipe. An example would be:
Baltimore Gas and Electric|Maryland
Delmarva Power|Maryland
local ACRRU runs are performed in main.py.
- Build unit tests for all application functions
- Build logging functionality for runtime performance monitoring
- Move input prompt structure from config file to separate text file for easier access/editing
- Integrate Database connections for caching ACRRU outputs
- Implement more tools ACRRU may use
- Integrate Self-Refinement post-processing
- Geographic query front-end for user-defined region rollup
- Integrate weighting scheme for entity contribution
- Collect geo-coded data on non-electric infrastucture providers
This repository uses Semanic Versioning.