Jewellery Quarter Bullion

AI Engineer Assessment

A timed coding challenge with a pre-built scaffold — function signatures, docstrings, and a mock API are all provided. You fill in the implementation across five sections: data pipeline, analysis, agent loop, prompt engineering, and risk scoring.

1–2 Hours

Duration

Python 3

Language

5 Sections

Parts

100 + 15 Bonus

Scoring

◆What You'll Build

›A bullion market intelligence agent — but you're not starting from scratch. A full code skeleton with function signatures, docstrings, and a working API mock is provided
›Fill in five interconnected modules: data pipeline, analysis engine, AI agent, prompt layer, and risk scoring — each with clear specifications and tests
›Work with a simulated precious-metals trading API and a mock LLM client that mimics real-world tool-calling patterns

◈The Challenge Structure

›Part 1 — Data Pipeline (12 pts): Fetch, normalise, deduplicate, and store price data from multiple endpoints with retry logic
›Part 2 — Analysis Engine (10 pts): VWAP, moving averages, anomaly detection, spread analysis, trend classification
›Part 3 — Agent Framework (40 pts): Build a ReAct-style agent that reasons via tool-calling loops against a mock LLM
›Part 4 — Prompt Engineering (18 pts): Craft prompts for tool selection and reasoning; parse structured output from noisy LLM responses
›Part 5 — Risk & Signals (20 pts): Risk scoring and trading signal generation combining your earlier analysis functions

⬡What We're Looking For

›Agent design and prompt engineering — the highest-weighted sections reflect what matters most in production AI systems
›Correctness — auto-scored tests are the primary metric; passing tests matters most
›Clean Python — type hints, docstrings, error handling, and readable structure
›Edge-case handling — empty data, zero volumes, missing values, API failures, malformed LLM output

◇Before You Begin

01The timer starts when you enter the challenge — have your environment ready

02You may use any reference material, documentation, or search the web

03You may use AI assistants (Copilot, ChatGPT, etc.) as you would at work — but you must understand every line you submit. Be ready to explain your code in interview

04Do not collaborate with anyone or share assessment content

05The session is monitored for context — tab switches and idle periods are logged

06Use the Run Tests button anytime — it does not count as a submission

07You can submit once only — no resubmission after final submit

08Your code is auto-scored against 49 tests across 5 categories

◇Scoring Breakdown

Part 1Data Pipeline

Part 2Analysis Engine

Part 3Agent Framework

Part 4Prompt Engineering

Part 5Risk & Signals

BonusCode Quality & Creativity

+15

Total

115

Admin access → Dashboard