Aleena

Aleena is an open-source lifecycle alignment agent for research software engineering collaborations. It turns meeting transcripts and chat conversations into structured, project-aware GitHub records so researchers and engineers can preserve decisions, surface risks, and keep implementation work aligned with project intent.

Aleena can:

• track terminology and flag conflicting definitions;
• extract meeting minutes, decisions, action items, risks, and open questions;
• create or update GitHub issues, discussions, comments, and draft pull requests;
• analyze existing GitHub activity for temporal alignment and missing context; and
• collect organization-wide evidence and generate evaluation reports.

Quickstart with Pixi

Install Pixi, then create your local configuration from the provided template:

cp .env.example .env

Fill in the required LiteLLM and GitHub credentials in .env, then use the task for the workflow you need:

# Build the frontend and run the Aleena web application
pixi run aleena-web

# Run the test suite
pixi run test

# Collect GitHub monitoring evidence
pixi run monitor --org uw-ssec

# Evaluate the latest monitoring snapshot
pixi run evaluate --action-items-target 5 --risks-target 2 --open-questions-target 2

Open http://127.0.0.1:8000/v2 after starting the web application. Pixi creates separate, reproducible environments for the application, tests, and monitoring workflows from pyproject.toml and pixi.lock.

For focused operational details, see the monitoring and evaluation guide and infrastructure guide.

System Architecture

flowchart LR
    A["Meeting transcripts and chats"] --> B["React + Vite frontend"]
    B --> C["FastAPI web application"]
    D["CLI / GitHub Action"] --> E["Services and pipelines"]
    C --> E
    E --> F["Alignment analyzers"]
    F <--> G["LiteLLM endpoint"]
    E --> H["GitHub integration"]
    H --> I["Issues, discussions, comments, and draft PRs"]
    I --> J["Monitoring collector"]
    J --> K["JSON snapshots"]
    K --> L["Evaluator and HTML report"]

Loading

• Frontend: The React and Vite interface accepts uploaded transcripts or pasted conversations, repository selection, analysis choices, and prompt review.
• Backend and services: FastAPI handles authentication and requests, while service and pipeline modules coordinate intake, analysis, state, and result delivery.
• Analysis layer: Specialized analyzers extract terminology, meeting outcomes, actionable work, and temporal alignment signals with the configured LiteLLM endpoint.
• GitHub integration: GitHub App and OAuth integrations discover repositories and create or update the collaboration artifacts produced by an analysis.
• Monitoring and evaluation: The monitoring pipeline collects GitHub evidence into versioned JSON snapshots, calculates deterministic metrics, and renders self-contained HTML reports.

Citation

The system design and motivation are described in Aleena: Alignment Agent for Research Software Engineering Collaborations. The work was presented at the KDD 2026 workshop AgenticSE @ KDD '26: Agentic Software Engineering (SE 3.0): The Rise of AI Teammates.

If you use Aleena in your work, please cite:

@misc{dani2026aleenaalignmentagentresearch,
  title         = {Aleena: Alignment Agent for Research Software Engineering Collaborations},
  author        = {Kshitij Dani, Cordero Core, Landung Setiawan, Carlos Garcia Jurado Suarez, Anshul Tambay, Vani Mandava, Anant Mittal},
  year          = {2026},
  eprint        = {2607.08043},
  archivePrefix = {arXiv},
  primaryClass  = {cs.SE},
  url           = {https://arxiv.org/abs/2607.08043}
}