Anya von Diessl signature
← Back to Projects

Project

LLM Agent with Tool Use, Memory, and Retrieval

DSPy ReAct Retrieval-Augmented Generation Vector Memory
PROJECTS.md — Portfolio (Workspace)
PORTFOLIO (WORKSPA…
MABOUT.md MEXPERIENCE.md MPROJECTS.md×
Portfolio src PROJECTS.md
1
# Projects
2
3
Selected engineering and machine-learning projects.
4
5
## Autonomous Agent Reliability & Assurance Platform
6
Adversarial evaluation framework scoring autonomous agent reliability across reasoning, tool use, and recovery.
8
9
## GPT-2 Fine-Tuning
10
Domain-adapted a pretrained GPT-2 model using PyTorch and Hugging Face Transformers.
12
13
## LLM Agent with Tool Use, Memory, and Retrieval
14
A ReAct-based agent with persistent memory and retrieval-augmented generation, built on DSPy.
⎇ main* ↓0 ↑3 ◯ 0 ⚠ 2 ⓘ 1 Ln 5, Col 1 Spaces: 2 UTF-8 LF {} Markdown ✓ oxc Token Count: 1,204 ✓ Prettier 🔔

Tools & technologies

DSPy ReAct Retrieval-Augmented Generation Vector Memory DSPy ReAct Retrieval-Augmented Generation Vector Memory

An agent that remembers, reasons, and reaches for the right tool.

A ReAct-based agent built on DSPy that pairs persistent vector memory with retrieval-augmented generation, enabling grounded, multi-step reasoning across conversations.

ReAct-based Reasoning architecture
Vector-based Persistent memory
Multi-step Task execution

Developed an intelligent LLM-powered agent capable of autonomous reasoning, tool invocation, persistent memory, and real-time information retrieval. Implemented a ReAct-based architecture using DSPy to dynamically orchestrate function calls, integrated vector-based memory for personalized multi-turn interactions, and engineered retrieval-augmented generation through web search and document parsing to improve factual accuracy on current information. Designed scalable backend workflows, recommendation algorithms, and state management systems to support reliable multi-step task execution while leveraging modern large language model frameworks for contextual decision making.