Alvin Vinod

Technical Assistant at University of Nottingham

Serving as a Subject Matter Expert (SME) supporting a Big Four firm on R&D tax audit assessments related to software engineering and AI/ML technologies. Reviewing technical claims, validating innovation narratives, and assessing technological uncertainties in complex software projects. Collaborating with engineering teams and auditors to produce clear, concise, and technically robust documentation.

Research Assistant - LLM Reasoning at University of Nottingham

Investigating differences between structured reasoning models (developed during my thesis work) and free-form reasoning models, with a focus on reliability, interpretability, and sample efficiency. Running early-stage experiments involving reasoning trace segmentation, sentence-level classification of chain-of-thought (CoT), and ablation studies to understand how different reasoning patterns emerge.

Research Assistant - LLM Post Training at University of Nottingham

Developed a system that uses inference-time compute methods to optimize the performance of LLMs on formula generation. The system can take a single/multiple text documents as input and create computation graphs / accurate formulas, with an average total generation time of < 1 minute, across queries with varying complexities. Optimized inference with SGLang, leveraging its reasoning parser, structured/JSON decoding, Radix Attention for prefix caching, tensor parallelism, and continuous batching to handle complex distributed queries efficiently.

Data Scientist at KPMG Global Services Private Limited

Developed a script to extract tables from over 1000 PDF invoices, performing data checks, leading to identifying & resolving inconsistencies in billing data. Containerised and set up inference and its end point monitoring. Implemented logic to incorporate table data from PDFs in an internal Document GPT application, which is a system that offers interaction capabilities with any PDF documents, fine-tuned specifically to extract ESG metrics from financial documents. Created core backend logic for an application automating the generation of over 500 resumes using predefined templates, incorporating NLP components to fix grammar and tense issues. Recognised with the 'Mega Star Award' for handling 5 complex projects, effectively engaging with demanding clients and ensuring project delivery.

Associate at ZoomRx Healthcare Technology Solutions Pvt Ltd

Engineered a harness to monitor and predict the effectiveness of promotional strategies using custom classifier models across 17 therapeutic areas, analyzing data from 4 US-based pharma and biotech clients, resulting in promotional strategies with a ~21% increase in recall. Developed a multi-channel engagement analytics pipeline that quantified COVID-19 disruption across 5 personal promotions and 14 non-personal promotion channels enabling data-driven customization of product offerings and accelerating campaign-impact monitoring by 40% for a large pharma company. Awarded 'Calmer of Storms' for project leadership.