# Applied AI prototypes, from crops to small models

- Period: 2023 – present
- Domain: Public sector, small businesses and independent work (Research and public sector)
- Role: AI developer, technical assistant and independent builder
- Key result: 25% → 75%, exact-match extraction accuracy after QLoRA fine-tuning of Llama 3.2
- Capabilities: Applied ML and research; AI evaluation and quality; Solution architecture and delivery
- Stack: Python, Computer vision, Llama 3.2, QLoRA, RAG
- URL: https://farhanbaluch.com/work/applied-ai-prototypes

Computer vision for crop disease and a policy chatbot for a federal department, AI prototypes for small businesses at the Vector Institute, and fine-tuning and RAG evaluation on my own time.

**AI angle.** The models underneath agents: vision, retrieval and fine-tuning. Knowing when a small fine-tuned model beats a large prompted one, and how to measure it, is what lets me choose the right tool on client work.

**What this shows:**
- I can fine-tune and evaluate models myself, not only call hosted APIs.
- I have published research: five peer-reviewed papers in fraud detection, computer vision and healthcare prediction.

### Work

- **Agriculture and Agri-Food Canada (2023).** Computer vision for early crop-disease detection, and a chatbot that answers farmers' policy questions.
- **Vector Institute (2024).** Helped small and mid-sized businesses take AI use cases from problem definition to a prototype in three months.
- **Small-model fine-tuning.** Fine-tuned Llama 3.2 with QLoRA on consumer hardware for structured data extraction. Exact-match accuracy went from about 25% to about 75%.
- **RAG evaluation lab.** An open workspace for benchmarking RAG pipelines before production, which cut manual evaluation time by 60%.
- **Research.** Five peer-reviewed papers in fraud detection, computer vision and healthcare prediction.