app/data/knowledge_base.md

Parsa Rouhi — Personal Knowledge Base

Personal Summary

My name is Amirparsa Rouhi (Parsa). I'm an AI/ML engineer with an MSc in Data Science & AI (Distinction, 2025) from Bournemouth University and a BSc in Computer Engineering. I'm currently based in the UK and actively seeking AI/ML engineering roles, open to relocation to London.

Education

• MSc Data Science & AI — Bournemouth University, 2025 (Distinction)
• BSc Computer Engineering — (completed before MSc)

Technical Skills

• Languages: Python (primary), SQL
• ML/DL Frameworks: PyTorch, TensorFlow, Hugging Face Transformers
• LLM & RAG: LangChain, LangGraph, RAG pipelines, pgvector, FAISS, LoRA/PEFT fine-tuning
• Backend: FastAPI, REST APIs
• Cloud & Deployment: AWS, Docker, Hugging Face Spaces
• NLP: Transformers, BERT, GPT-based models, Swin Transformer, sequence modeling
• Computer Vision: Medical image analysis, radiology report generation

Key Projects

RADSAM Platform

A production-grade radiology AI platform. Built an automated radiology report generation system using a Swin Transformer encoder paired with a Cerebras-GPT-1.3B decoder. Evaluated on the IU-Xray dataset. The project became the foundation for my MSc dissertation and a conference paper submission to IUI 2026.

R2GenTransformer (MSc Dissertation)

"A Lightweight Transformer Framework for Automated Radiology Report Generation." Focused on prompt engineering techniques with a Swin encoder + Cerebras-GPT-1.3B architecture on the IU-Xray dataset. Submitted to IUI 2026 conference.

ParsaGPT

A production-deployed conversational AI system. Built end-to-end with a custom RAG pipeline, demonstrating real-world LLM deployment skills.

Multi-Agent RL Trading System

Built a multi-agent reinforcement learning trading system using PPO (Proximal Policy Optimization). Explored deep RL for financial market modeling.

Kaggle — Predictiva Competition (Pairwise Trading Agent Prediction)

Progressed from a 57% baseline to ~89% validation accuracy through:

• LSTM sequence modeling
• Feature engineering
• Deep learning ensembles (feedforward neural networks + rank-based ensembling)

Cryptocurrency Token Pricing (Academic Research)

Research project modeling Nash equilibrium dynamics and behavioral economics for token pricing. Used game-theoretic approaches. Targeting academic publication.

Career Goals

• Seeking AI/ML Engineering roles in the UK (junior to mid-level)
• Open to industries: FinTech, HealthTech, EdTech, biotech, general AI product companies
• Requires visa sponsorship for future work authorization
• Preference for remote-friendly or London-based roles
• Particularly interested in: LLM applications, RAG systems, production ML, NLP

What Makes Me Stand Out

1. Production experience: RADSAM and ParsaGPT are deployed systems, not just academic exercises
2. End-to-end ML: I've done research (IUI 2026 paper), built systems, and deployed them
3. Breadth + depth: Medical AI, financial AI, LLMs, RL — across multiple domains
4. Distinction-level MSc: Graduated with top marks from a UK university
5. Hands-on LLM expertise: LoRA fine-tuning, RAG pipelines, multi-tool agents (LangGraph)

Contact & Profiles

• Open to being contacted by recruiters
• GitHub: visible with active project contributions
• LinkedIn: active profile
• Based in UK, eligible to work (requires future sponsorship)

Personality & Work Style

• I enjoy bridging rigorous academic methods with practical, deployable systems
• Strong interest in financial markets and AI applications in trading
• Self-driven learner — completed competitive Kaggle challenges alongside MSc
• Comfortable working in fast-paced, research-oriented environments