RTS Questionnaire · Question 15 of 52 · file: answers/q15.html
Have you worked with Generative AI, Agentic AI, or MCP (Model Context Protocol)? Good to have.
GenAIAgentic AIMCP
✅ Yes — GenAI, agentic workflows and MCP are an active part of my current engineering practice, not just theoretical knowledge.
🧪 Interactive proof: this answer has its own page with a live GenAI workflow, the RAG pipeline and MCP server behind it and the acceptance criteria — open the Q15 GenAI proof →
Technical detail
- 🧠 Generative AI delivery — RAG pipelines (chunking strategies, embeddings, vector search with FAISS/pgvector-style stores), prompt design and versioning, context-window and cost management, response guardrails, and evaluation with metrics such as EM/F1/ROUGE plus human review.
- 🛠️ Model fine-tuning — parameter-efficient training (4-bit quantized QLoRA adapters on Llama-3-class models) over domain datasets, with loss tracking and adapters served behind a FastAPI inference API (
/askstyle endpoints). - 🤝 Provider integration — OpenAI APIs, Gemini, Claude and local/open-source models behind a common abstraction so a project can switch provider or fall back without rewriting business logic.
- 🤖 Agentic AI — durable tool-calling workflows where agents plan, call APIs/databases, validate outputs and retry safely; idempotent side effects, structured tool schemas, human-in-the-loop approval gates, tracing and cost/rate-limit control.
- 🔌 MCP (Model Context Protocol) — the client/server model for exposing tools, resources and prompts to LLM hosts; designing MCP servers that wrap internal systems with least-privilege access, typed schemas and audit logging, so assistants reach enterprise data in a governed way.
🔐 Governance is part of the delivery: PII handling and anonymisation, least-privilege credentials, prompt-injection awareness, and observability over token usage.
🧾 CV evidence: 2025 CV: AI-assisted delivery and 'durable agentic workflows'; technical assessment with Llama-3 QLoRA fine-tuning, embeddings + FAISS RAG, FastAPI serving; AI/ML and GenAI platform documentation in the assessment record.