Production-minded LLM samples you can clone and run offline: retrieval with cited sources, agents with a human approval gate before writes, and safe LLM-to-SQL via vetted query templates. Built as public portfolio demos aligned with real Upwork / consulting work—not client code, no real data, no committed API keys.
| Demo | Stack | What it shows |
|---|---|---|
rag-citations |
TypeScript / Node | Ingest → embeddings → top-k retrieve → answer with source snippets + eval set |
agent-approvals |
TypeScript / Node | Tool-calling agent; reads auto-run, writes require y/n; JSONL audit log |
llm-sql-safe |
Python / SQLite | LLM picks from vetted SQL templates (never free SQL); replayable logs |
Nabil Ashraf, CTO @ XCatalyst
GitHub: @nabilashraf · Org: @chumchumagency (branded XCatalyst)
These are sanitized demos for portfolio and interview use. They are not production client systems and contain no client data.
git clone https://github.com/nabilashraf/ai-demos.git
cd ai-demos
Each demo is self-contained. Mock / offline mode is the default (no OPENAI_API_KEY needed). Set the key in a local .env only if you want real model calls.
cd rag-citations
npm install
npm run ingest
npm start -- "What is the refund window for Pro plans?"
npm run eval
# optional API: npm run api
cd agent-approvals
npm install
npm start
# follow prompts: propose → approve/deny write tools
# smoke: npm start -- --auto-script --approve
cd llm-sql-safe
python3 -m venv .venv
./.venv/bin/pip install -r requirements.txt
./.venv/bin/python -m src.seed
./.venv/bin/python -m src.cli "What was AAPL closing price on the last trade day?"
MIT — see LICENSE.