Interview Prep — AI Specialist (Digital Twin)
This folder prepares you for every question category you will face interviewing for an AI Specialist role on an Operational Digital Twin programme (Parsons, Jacobs, Leidos, AECOM, or equivalent GovTech / national-infrastructure clients).
Files
| File | What it covers |
|---|---|
| 01-concepts-cheatsheet.md | 25 crisp answers to the questions every panel will ask |
| 02-system-design-walkthroughs.md | 4 practice prompts with timed guidance |
| 03-behavioral-questions.md | STAR stories for GovTech / secure-environment / multinational contexts |
Interview Format to Expect
| Round | Duration | What They Probe |
|---|---|---|
| Screening (HR) | 30 min | Motivation, location/travel, English fluency, clearance eligibility |
| Technical Screen (Lead) | 60 min | LLM fundamentals, RAG, fine-tuning concepts, air-gap awareness |
| Technical Deep Dive | 90 min | System design + live coding / whiteboard (Labs 1–4 territory) |
| Domain Fit (Programme Manager) | 45 min | GovTech / infrastructure awareness, stakeholder comms, governance |
| Behavioural (Panel) | 45 min | STAR stories; teamwork in multinational contexts; security mindset |
Fastest Path to Readiness
- Do the labs first — answers without artefacts land weakly. The labs give you concrete numbers ("84% retrieval accuracy at chunk size 1000", "TTFT 905ms warm").
- Memorise the 30-second mental models from each HITCHHIKERS-GUIDE.md. Those models are the answer scaffold for 80% of technical questions.
- Read the JD line by line (../jd.md) and map each requirement to a lab or concept.
- Practice the four system design prompts out loud, timing yourself at 45 minutes each.
- Prepare three STAR stories and flex them to the question variants in file 03.
Domain Context You Must Know
The Parsons JD describes a Geospatial Digital Twin for a national infrastructure owner in the Middle East. This means:
- Zero internet at runtime — no OpenAI, no Hugging Face inference, no pip install.
- Arabic language — explicitly mentioned as a strong advantage. Know multilingual embedding models and multilingual LLMs (Jais-30B, AceGPT, Qwen2.5 multilingual).
- SCADA / IoT data — the AI assistant queries real-time sensor feeds. Understand SCADA architecture at a level that lets you have an informed conversation.
- Geospatial data — GIS formats (GeoJSON, Shapefile, WFS/WMS), coordinate systems (WGS84, UTM), spatial queries. You don't need to be a GIS expert but you must not be confused by the terms.
- Data governance / security classification — national infrastructure data is sensitive. The JD mentions "compliance with security and data-governance requirements". Know what that implies: access logging, data minimisation, no model exfiltration, classification markings in prompts.
- Stakeholder communication — you will explain hallucination risk, confidence scores, and model limitations to non-technical infrastructure engineers and government officials. Practice plain-language explanations.