shayne williams

shayne williams

Applied AI Engineer and Data Scientist
UAE
English
Contract Hybrid

About Me

Applied AI Engineer and Data Scientist with 5+ years’ experience spanning agentic AI and RAG systems at Blackford and Worknest, and enterprise machine learning and MLOps at PwC and Tata Consultancy Services — including a…

Experience

AI Engineer

Blackford · Abu Dhabi , Abu Dhabi , United Arab Emirates (UAE)
Jul 2025 - Sep 2026 · 1 year 2 months

<p>Built the ontology-construction and digital-twin simulation layer of a multi-agent decision-intelligence platform: agents that reshape source tables into typed business objects, classify them against domain ontologies (OSDU/FIBO/CIM/GS1), and mine functions and actions from documents — validated end-to-end on synthetic data modeled on a target client’s schema.</p><p> Built a typed intermediate representation so agents propose transformation/mapping plans, executed by a deterministic compiler as allowlisted PySpark operations, preventing arbitrary code execution.</p><p> Engineered cross-source entity resolution and Qdrant-based semantic retrieval connecting source columns, terminology and ontology concepts, abstaining on ambiguous matches to preserve auditability.</p><p> Modeled a five-part Action Registry (signature, preconditions, effects, policy, handler) and a Rete-style rule engine with stratified negation and truth maintenance, combined with Monte Carlo sampling to quantify outcome ranges under proposed actions.</p><p> Built an event-sourced, copy-on-write branching state model — an immutable Iceberg fact log as the system of record, backing a fast Postgres-cached read layer — so forecast, what-if and counterfactual scenarios never mutated the baseline graph.</p><p> Built a Kubernetes-hosted microservice deployment: independently deployable services exposed as MCP-compatible tools over FastAPI, decoupling the layer from whichever upstream agent-orchestration framework consumed it.</p><p> Built an ontology-grounded RAG query-copilot agent atop the layer as a LangGraph orchestration of compiled subgraphs and plain chains, resolving questions into linked entities and intents and routing across knowledge-graph lookup, corpus retrieval and the simulation engine, with citation-grounded verification before response synthesis.</p><p> Designed the self-hosted inference plan for the air-gapped target environment: separate interactive and batch vLLM pools on Kubernetes, LoRA-served rule-extraction adapters, prefix caching, and quantised variants gated by the golden eval set, with replica counts sized from benchmarked operating points.</p>

ML Engineer

Worknest · London , England , United Kingdom (UK)
Jul 2024 - Jun 2025 · 11 months

<p>Built and deployed a bounded multi-agent Employment Case Resolution Copilot — FastAPI and LangGraph on Azure Container Apps, VNet-isolated with Entra ID authorisation — with specialist agents for triage, evidence retrieval, policy lookup, drafting and validation, governed by Unity Catalog for PII pseudonymisation, access, lineage and auditability.</p><p> Engineered the case pipeline: document/email ingestion, OCR, PII detection, case-scoped pseudonymisation, and quarantining of content failing safety checks; built hierarchical chunking preserving structure and metadata for citation.</p><p> Built hybrid evidence retrieval — BM25 plus Qdrant semantic search, reranked by a fine-tuned cross-encoder on pseudonymised query-evidence pairs — to prioritise citation-supporting chunks before answer generation.</p><p> Built an agent router combining few-shot intent classification with deterministic backend policy validation; resolved references via exact ID/email/name/alias matching before LLM fallback; enforced human adviser approval for legal, external or consequential actions.</p><p> Created pseudonymised golden evaluation trajectories and GitHub Actions CI regression gates — hosted runners for unit tests, ephemeral private Azure runners for live-endpoint tests — covering routing, tool-use validity, citation support and safe escalation, with LangSmith observability throughout.</p><p> Engineered a company knowledge graph and Copilot for cross-sell insights; delivered the company’s top dashboard tracking contracts, financials and operations.</p>

AI Specialist

PwC · London , England , United Kingdom (UK)
Sep 2022 - Jul 2024 · 1 year 9 months

<p>Led development and production deployment of a full MLOps pipeline in Databricks for proactive and diagnostic technology-equipment analytics, processing device telemetry at scale with PySpark and MLflow for reliable, auditable delivery.</p><p> Built LSTM time-series models to predict which machines were likely to fail or need maintenance in the following week, enabling targeted monitoring and proactive intervention after software updates and rollouts.</p><p> Built CatBoost models for BSOD, PC-performance and application-performance diagnostics; created SHAP-based dashboards guiding IT operations teams on scheduling maintenance, repairs and remediation across the PwC device fleet.</p>

Data Scientist

Tata Consultancy Services · London , England , United Kingdom (UK)
Sep 2021 - Aug 2022 · 11 months

<p>Built a 360° HR analytics product (churn/skills-gap forecasting, NLP candidate profiling); Azure ML Studio pipelines with Data Factory.</p>

Skills

Python PySpark Java SQL LangChain LangGraph LangSmith Agentic AI Multi-Agent Orchestration Model Context Protocol (MCP) Generative AI RAG LLMs Transformers NLP LLM Inference LLM Serving KV-cache Memory Continuous Batching Quantisation GPTQ AWQ FP8 Benchmarking Capacity Planning Knowledge Graphs Ontologies Digital Twin Simulation Machine Learning Deep Learning CNN RNN LSTM GNN MLOps MLflow Model Deployment Databricks Spark Apache Iceberg PostgreSQL Qdrant Azure Azure Container Apps Azure Application Insights OpenTelemetry LaunchDarkly Kubernetes Data Pipelines Entity Resolution
Report this Profile?