Ratneshwaran Maheswaran

Research Assistant · UCL Centre for Artificial Intelligence

I am a Research Assistant at the UCL Centre for Artificial Intelligence, working on reliable, evaluation-led LLM systems for high-stakes settings. My current work — in collaboration with Freedom from Torture — prototypes a safety-aware multilingual LLM voice assistant, with evaluation harnesses, risk-monitoring, and audit-friendly logging.

Previously, I worked at the Secrier Lab at the UCL Genetics Institute, building reproducible pipelines for large-scale single-cell datasets and benchmarking foundation-model approaches under batch effects and dataset shift. I hold an MSc in AI for Biomedicine and Healthcare from UCL and a BEng in Electronic and Computer Engineering from the University of Nottingham, where I received the Vice-Chancellor's Medal (top 0.1%).

My interests lie in reliable and safe LLM systems, AI evaluation and benchmarking, agentic AI and workflow reasoning, reproducible ML, and AI for scientific and biomedical data.

News

Mar 2026 Teaching Assistant for COMP0190 (AI for Domain Specific Applications Project) at UCL Computer Science.
Dec 2025 Completed MSc in Artificial Intelligence for Biomedicine and Healthcare at UCL with Distinction.
Oct 2025 Joined the UCL Centre for AI as a Research Assistant, working on a safety-aware LLM voice assistant with Freedom from Torture.
Sep 2025 Completed work at Secrier Lab on foundation-model benchmarking for single-cell data.
2024 Awarded the Vice-Chancellor's Medal at the University of Nottingham (top 0.1%) for academic, authorial, and leadership contributions.

Research Interests

Reliable and safe LLM systems — evaluation, monitoring, and risk-aware design for high-stakes settings.
AI evaluation and benchmarking — experimental design, error analysis, and transparent iteration.
Agentic AI and workflow reasoning — reasoning, planning, and tool use with LLM-based agents; novel metrics for LLM-derived workflows.
Reproducible machine learning — research software and pipelines that support reuse and clean iteration.
AI for scientific and biomedical data — single-cell genomics, cross-dataset generalisation, clinical NLP.

Selected Projects

Grounded By Design: safety-aware LLM voice assistant (prototype).

R. Maheswaran. UCL Centre for AI, in collaboration with Freedom from Torture, 2025–2026.

Research prototype exploring reliability and safety mechanisms for multilingual voice-based LLM support in sensitive contexts. Implemented monitoring and guardrails to detect high-risk content and route to conservative handover pathways; end-to-end STT and TTS pipeline with instrumentation for auditability (logging, transcript capture, outcome signals). Manuscript in preparation.
Record2Table: multi-document processing pipeline for structured outputs.

R. Maheswaran, 2026.

Pipeline that ingests multiple documents per case and produces consolidated structured JSON, tables, and concise summaries. Schema-based extraction, normalisation utilities, provenance-friendly outputs with source snippets, and evaluation scripts.
Zero-shot cell annotation with foundation models.

R. Maheswaran. Secrier Lab, UCL Genetics Institute, 2025.

Zero-shot pipelines for cell classification and multi-class state prediction across multiple datasets. Evaluation with Accuracy, F1, AUROC, and AUPRC alongside diagnostic visualisations for generalisation and dataset shift. Blog post; LangCell and Geneformer on GitHub.
LitAgent: AI agent for literature search and synthesis.

R. Maheswaran. Open-source system, 2025.

LangGraph-driven multi-agent system for automated search and synthesis across arXiv and PubMed with Crossref enrichment. Exports structured outputs (Markdown, JSON, CSV), supports CLI and REST API workflows, and includes critique checks to flag overclaims.
Enhancing aspect-based sentiment analysis with adversarial training and hierarchical aggregation (BERT).

R. Maheswaran, 2025.

Improved aspect extraction and sentiment classification by combining adversarial training with hierarchical aggregation.

Experience

Mar 2026–: Teaching Assistant, COMP0190. UCL Computer Science. AI for Domain Specific Applications Project (part-time).
Oct 2025–: Research Assistant. UCL Centre for AI, in collaboration with Freedom from Torture. Safety-aware LLM voice assistant; evaluation harnesses and risk-monitoring.
Mar–Sep 2025: MSc AI Researcher. Secrier Lab, UCL Genetics Institute. Reproducible pipelines and foundation-model benchmarking on single-cell data.
2022–24: President, Faculty of Engineering Society (OPAD). Open-Source Assistive Devices, University of Nottingham. Directed 12+ student-led projects; secured £1,500 in grants.
2020–: Founder & CEO, 40seconds.org (Ratneshwaran Foundation). Non-profit on youth mental health and human rights; 120+ volunteers across 19 countries. Currently on hiatus.

Education

2024–25: MSc, Artificial Intelligence for Biomedicine and Healthcare. University College London. Distinction.
2021–24: BEng (Hons), Electronic and Computer Engineering. University of Nottingham.

Teaching

COMP0190 — AI for Domain Specific Applications Project, UCL (MSc). Teaching Assistant, 2026–present.
Peer mentoring & OPAD project supervision, University of Nottingham, 2022–24.

Honours & Awards

Vice-Chancellor's Medal, University of Nottingham (2024) — top 0.1%, for contributions as author, philanthropist, and student leader.
EEE Undergraduate Studies Award, University of Nottingham (2024).
Nottingham Advantage Awards — Student Leader (2024); OPAD & Peer Mentor (2023).
Union Prize & Student Representative of the Year, University of Nottingham (2022).
Top 6 Teen Heroes, India (2021).
Invited speaker — World Summit AI, TEDx, university conferences.

Contact

The best way to reach me is by email at ratneshwaran.maheswaran.21@ucl.ac.uk. I'm always happy to hear from students, collaborators, and anyone working on trustworthy AI for healthcare.