<aside> 🗣 Where I share my thoughts and projects within the realm of ‘AI and Automation’. My projects span 1.5 years of building LLM’s and agentic solutions. If you’d like to see a specific solution built, drop me a note.

</aside>

$$ \Large\text{Portfolio Project: The Persona-Forge} $$

$$ \Large\text{Portfolio Project: A project fusing my interest in psychology with a selection of popular AI frameworks, to demonstrate the ‘Art of the Possible’.} $$

The Persona-Forge project is built upon similar building blocks of modern AI 'Code-Generators', repurposed to demonstrate how AI can be applied to any field. It's designed as an 'art of the possible' demo to help therapists and mental health hospitals modernize and care for patients.

It utilizes automation to enhance clinicians' delivery of progress notes and analysis. The workflows are all built in Langgraph, a production-grade agentic process flow library, utilizing local models to reduce costs and comply with data security regulations.

Features include:

• My customization of the LangGraph open source deep agents library, allowing a team of 8 Agents with 6 models (One main agent can spawn infinite subagents) working on their own computer (E2B replacing LangGraph state engine) to retrieve data and analysis from a variety of sources (Knowledge Graph, Embeddings, Pubmed, Web), compiling a ‘Therapy Progress Note’ with data points, made possible by AI.
• A robust data-tagging workflow to analyze therapy sessions using 7 psychological frameworks. This is used to supply an LLM with a script of 24k tokens, the data is then processed and analyzed by the LLM to generate comprehensive insights and actionable recommendations.
• A knowledge graph (KG) workflow to store and query the therapy session data, including the psychological analysis results.
• A voice-enabled chat interface (Python, FastAPI, Gradio) providing a user-friendly interface for interactive sessions, enabling users to query the therapy session data and insights.
• A conversational text interface (built with ‘Copilotkit & AG_UI) allowing users to interact with the data and gain insights through natural language queries, enabling a seamless and intuitive user experience. The chatbot uses 'Hybrid Graph-RAG' to query the KG, using the psychological analysis results and insights gained from the tagging workflow.
• A comprehensive dashboard for monitoring the progress of the therapy sessions (text or audio), with ‘Just in Time’ (Web layer) visualizations of the psychological analysis results.

1. Features Overview (Sound on)

https://youtu.be/Ny0H456Frr8

2. Psychology Frameworks Analysis

https://youtu.be/NklmW2PSNxU

Workflow 1 - Psychology Frameworks

The Psychology workflow (completed prior to the deep agents one) for my Persona-Forge. It wasn't really designed for demoing but I did my best to display it. First Workflow: The agent parses all QA Pairs, tags against 7 psychology frameworks. Second Workflow: Agent parses through output and creates embeddings and Cypher (knowledge graph language) for Hybrid-Graph-Rag.

Simplified outline of the Deep Agent workflow.

3. Therapy Note Creation

https://youtu.be/1di48I3zs5k

Deep Agent Workflow: Agents/LLM's used:

Primary: GPT-OSS:20b - LM Studio Articulate Agent: GPT-OSS:20b - Ollama (via networked PC) Context Agent (condenses large blocks of text) : Either Gemini 2.5 (where anononmized data) or Granite4:27b (local) Nimble Task Agents: Nemotron Nano 9B or Granite4 Tiny

View Project on Github (Tutorial Videos to Appear Here Shortly)

https://github.com/David-Barnes-Data-Imaginations/persona-forge-langgraph

Use this if the link above breaks: persona-forge-langgraph

My Data Presentations from LSE: → https://github.com/David-Barnes-Data-Imaginations/LSE-Data-Analytics-Assignments

For Recruiters: My Skills

Various other projects & work from my LSE course can also be found on my Github profile

<aside> 🗣

Those looking for my langgraph deep agents e2b customization, it’s here (has a Claude Code style CLI UI added)

</aside>

My Technical Blog - Where I discuss Model Training Math, Research Papers & other bits

My Non-Technical Blog

My Story

👋 I’m David - a Senior HR leader transitioning to AI Engineering. Bringing 15 years of process improvement expertise to AI implementation, following a career change. I have a degree in Software Engineering, and recently studied Data Analytics for a year at LSE in London. I finished my course in summer 2024 with a hasty pivot towards AI, after language models were unleashed on the world whilst I studied.

I was inspired by the self-directed learning paths of people like OpenAI/Anthropic co-founders Tom Brown and Greg Brockman, and took a similar approach - committing to intensive, project-based learning in AI rather than taking a tangential role that would only give me peripheral exposure.

At this point I struggle to think of any architecture, system or endeavor in this space that I haven’t had experience of, but the one’s I have include:

• Building multi-agent workflow with frameworks - Langchain, LLama-Index, SmolAgents.
• Advanced recall with Hybrid-Graph-RAG, Context Management and Task Lists.
• User interfaces ranging from command line to dynamic generation.
• Automation of complex processes.
• Model training and fine-tuning.
• Sandboxing with E2B Linux Kernels for safe agent use during workflows
• Cloud computing in GCP & Azure

I’ve also completely lost track of when I’m supposed to use an ‘S’ or a ‘Z’ when writing English, so I reserve the right to use them interchangeably…