Published by the Government Operational Research Service on 23 September 2026, Vlad’s career profile does more than sketch one analyst’s route into government. Read alongside Ministry of Justice transparency material, it sets out how operational research, data science and product management are being combined inside probation and justice services, with ethics, risk control and documentation treated as routine delivery work rather than add-ons. (gov.uk) The profile places Vlad in the Ministry of Justice as an Associate Data Science Product Manager in Probation Data Science, following earlier work as a Senior Data Scientist on data linking, fines enforcement, reconviction rate estimation, employee insight work and a labour market dashboard. That mix of work shows a department using analytical staff across both service operations and evidence production. (gov.uk)
His route into government began with a BSc in Security and Crime Science at University College London, combining criminology, social research, psychology and coding, followed by a six-month placement in the Mayor’s Office for Policing and Crime evaluation unit. In the case study, he says operational research only became visible after he joined the Civil Service, when he recognised that evaluation of crime prevention measures already drew on the same habits of structured analysis. (gov.uk) That background fits the way GORS describes its own remit. The profession says it supports policy-making, strategy and operations across government, has more than 1,200 analysts, and usually recruits from highly numerate disciplines; official recruitment guidance says candidates are assessed through Civil Service Success Profiles and the GORS technical framework. (gov.uk)
In the profile, Vlad describes his current role as covering product vision, ethics, stakeholder management, risk identification and documentation for data science projects in probation, with current work focused on large language models. He presents technical literacy as useful not only for model development but for explaining limitations and caveats to non-technical audiences. (gov.uk) For policy readers, that is the clearest signal in the piece. The account presents public-sector AI as a managed product function inside an existing service, not as a stand-alone research exercise. That reading is an inference from the responsibilities set out in the case study and the supporting transparency records. (gov.uk)
The most immediate example is the Contact Log Semantic Search tool. Vlad says probation staff write millions of contact reports each year and that the notes are unstructured, numerous and written in different ways, making it difficult to retrieve the right material when staff need it for case preparation, handover or risk assessment. The Ministry of Justice’s algorithmic transparency record describes the same tool as a way of helping practitioners search large volumes of Delius contact text more efficiently. (gov.uk) The technical design is more specific than the general term AI might suggest. According to the transparency record, the service combines literal search using Okapi BM25 with semantic search based on Mixedbread AI’s mxbai-embed-large-v1 model; the semantic element only sees the free-text notes field, returns results inside nDelius, and allows users to order those results by relevance or date. In March 2025, the department recorded around 1,000 daily users and an average of 5,000 searches each weekday. (gov.uk)
The governance arrangements are set out in equally plain terms. The transparency record says the tool does not replace professional judgement: probation practitioners decide what to search for, read the returned contacts themselves, and can bypass the search tool altogether in favour of manual review. MoJ also says literal matches are retained, semantic search only adds further results, user activity is logged securely, and the model is subject to ongoing monitoring and a formal annual review. (gov.uk) That makes the benefit claim narrow but operationally useful. The Justice AI Unit says semantic search is intended to reduce repeated search attempts, improve access to relevant information and free staff time for higher-value work such as risk management, while the transparency record says testing did not find differences that consistently or disproportionately affected a single demographic group. (ai.justice.gov.uk)
The second project highlighted in the profile concerns person data linking across the criminal justice system. Vlad describes the basic problem in administrative terms: records are messy, identifiers are not uniform, and a single person may appear under different names, addresses or system IDs. In his account, he worked on enhancing an existing Splink person-linkage output by adding another data source, and he notes that the open-source package has recorded more than 10 million downloads globally. (gov.uk) MoJ’s Data First transparency record explains why such linkage work is important. The department says no reliable common identifier exists across many justice systems, that previous open-source tools were not suitable at the required scale of tens to hundreds of millions of records, and that deterministic rules-based linking created too many missed matches for complex datasets. Splink was built to assign probability scores, deduplicate records within systems and generate linked identifiers across domains including courts, prisons and probation. (gov.uk)
The distinction between operational use and research use is important. MoJ says the Data First linked datasets are anonymised, are provided to accredited researchers to study trends across the justice system, and are not intended for operational decision-making; the department also says releases go through manual sample review, internal pipelines run weekly, and the programme had facilitated around 40 research projects at the time of publication. (gov.uk) The closing advice in Vlad’s profile is comparatively simple: analytical aptitude, logical thinking and clear communication matter more than attachment to one coding language. Set against GORS guidance on numerate entry requirements and formal assessment, the profile ends up serving a second function beyond recruitment. It gives policy readers a concrete picture of how AI search and record linkage are being built, reviewed and bounded inside day-to-day justice administration. The final sentence is an inference based on the cited sources. (gov.uk)