DLens

DeepLense AI Scientist

Agentic AI for autonomous scientific workflows in gravitational lensing research

Technical Documentation

Agent Tracker

Overview

DLens is a multi-agent framework for autonomously orchestrating scientific workflows in gravitational lensing research. It provides agentic capabilities for reasoning, decision-making, and workflow automation, enabling LLM-powered agents to coordinate complex multi-step processes with minimal human supervision.

Technical Foundation

DLens is built as an abstraction layer over Pydantic AI, leveraging its type-safe agent framework to create robust, composable agentic workflows. This foundation provides structured data validation, seamless integration with language models, and a clean API for building complex multi-agent systems.

Design Guidelines

DLens is designed to operate primarily with locally hosted or on-premise language models, avoiding dependencies on third-party model services. System prompts are constrained, focused, and information-dense, minimizing context usage to reduce both computational overhead and the risk of hallucination.

DLens follows a divide-and-conquer approach to agent design. Complex tasks are decomposed into smaller, well-scoped subproblems, each handled by a dedicated agent or a single LLM call. This improves reliability, reduces cognitive load on individual models, and enables more predictable and debuggable workflows.

For complex multi-tool-use agents that require larger LLMs with API access, DLens also supports a ReAct (Reasoning + Acting) framework, enabling iterative reasoning-action loops where agents can plan, execute tools, observe results, and adapt their strategy across multiple steps.

Quick Start

This guide walks you through setting up the dlens library.

1. Install uv

uv is a fast Python package and environment manager used by this project.

macOS / Linux

curl -LsSf https://astral.sh/uv/install.sh | sh

Windows

Do yourself a favor and switch to Linux or macOS.

(But if you insist: powershell -c "irm https://astral.sh/uv/install.ps1 | iex")

Restart your terminal, then verify:

uv --version

2. Clone the repository

git clone https://github.com/pranath-reddy/DeepLense-AI-Scientist.git
cd DeepLense-AI-Scientist

3. Create a virtual environment

uv venv
source .venv/bin/activate  # macOS/Linux
# .venv\Scripts\activate   # Windows

4. Install the package and dependencies

uv pip install -e .

5. Run the sanity check or pytest

uv run python scripts/sanity_check.py

Or run the test suite:

uv run pytest