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