Summary

Mission

Design, build, and own a working prototype that detects emerging threats to stablecoin peg stability, forecasts depeg risk, ingests and interprets compliance/regulatory context, fuses heterogeneous signals into a composite risk index (SPRI), and surfaces explainable, prioritised alerts to risk and compliance stakeholders. This is a hands-on role combining system architecture, data/model engineering, and operational decision-support delivery.

Key Responsibilities

1. Define and own the end-to-end PoC architecture: signal ingestion, feature engineering, anomaly detection, depeg probability forecasting, NLP compliance/context pipeline, signal fusion, explainability, alerting, and briefing generation.
2. Translate business risk objectives (peg integrity, regulatory drift, depegging probability) into concrete data contracts, JSON schema definitions, model interfaces, and alert logic.
3. Build and operate real-time and batch pipelines ingesting heterogeneous data: market/peg dynamics, order book/depth, on-chain flows (holder entropy, net imbalances, graph structure), reserve/backing health, regulatory text, and external sentiment signals.
4. Develop and tune AI models:
5. Unsupervised/hybrid anomaly detection for behavioural irregularities.
6. Supervised short-horizon depeg probability forecasting with calibration.
7. NLP components for topic classification, semantic drift/change detection in policy text, and sentiment/concern extraction.
8. Fuse signals into a composite risk index (SPRI) and design prioritisation logic that amplifies correlated stress (e.g., behavioral anomaly + regulatory shock).
9. Implement explainability (feature attribution, templated rationales) so alerts are transparent and actionable for compliance/risk analysts.
10. Build or shepherd a lightweight triage interface (dashboard or API) exposing alerts, scores, explanations, and recommended next steps.
11. Create and curate ground truth via historical and synthetic stress/depeg scenarios; drive evaluation metrics (detection lead time, calibration, alert lift).
12. Embed a human-in-the-loop feedback loop: capture analyst annotations (accept/reject, priority adjustment) and iteratively refine models, fusion weights, and thresholds.
13. Prepare and deliver stakeholder-ready incident walkthroughs and briefing summaries; own the final PoC report with quantitative findings and productionisation recommendations.
14. Ensure secure, auditable data handling, and that AI outputs are framed as decision support with human oversight.

Required Qualifications

1. 4+ years of hands-on experience building and deploying applied AI/ML systems, including both architectural design and implementation (anomaly detection, time-series forecasting, NLP).
2. Strong Python expertise and familiarity with core libraries (scikit-learn, PyTorch/TensorFlow, Hugging Face transformers, SHAP or equivalent explainability tools).
3. Proven ability to ingest and engineer features from heterogeneous sources (numerical time series, graph/flow data, unstructured text) and align them temporally for fusion.
4. Experience designing and combining multiple risk signals into composite indices or scoring systems; comfortable with both expert-driven and learned fusion logic.
5. Practical knowledge of NLP for document classification, semantic similarity/drift detection, and sentiment extraction.
6. Track record of building human-in-the-loop feedback mechanisms to improve model quality iteratively.
7. Strong product/operational instincts: able to turn model outputs into alerts, rationale, and concise briefings for non-technical stakeholders.
8. Excellent communication and collaboration—can work directly with compliance/risk SMEs and present to senior stakeholders.
9. Must be based in the UK (due to regulatory engagement and coordination with UK-centric stakeholders).

Success Metrics (for the PoC)

1. Reliable anomaly detection and depeg probability forecasting on synthetic/historical events with measurable lead-time advantage.
2. Composite SPRI that meaningfully prioritises genuine risk incidents over noise.
3. Alerts consistently enriched with contextual compliance/regulatory signals and accompanied by clear explainability.
4. Analyst feedback loop functioning: measurable improvement in precision/recall or alert utility over iterations.
5. Stakeholder-ready demo delivering end-to-end incident narratives and a concise recommendation report.

Engagement Details

1. Duration: 6-month prototype engagement (with strong possibility to transition toward a production strategy/extended role).
2. Reporting: Direct to PoC sponsor (CTO / Head of Risk); responsible for biweekly demos.
3. Location requirement: UK-based

Application Materials Requested

1. Brief case study or examples of similar systems (anomaly detection + compliance signal fusion, risk scoring, NLP drift detection).
2. High-level sketch of how you would architect fusion of behavioural and regulatory signals into explainable alerts.
3. References or prior work in regulated/financial contexts if available.

Skills

• Machine Learning
• Python