Jasivan Sivakumar

I

About me

I am a PhD candidate in Natural Language Processing (NLP) at the University of Sheffield. I started at the Speech and Language Technology (SLT) Centre for Doctoral training (CDT) in September 2021. My focus lies in numerical reasoning especially how to evaluate and improve smaller accessible general-prupose model to enhance their numerical understanding in downstream tasks. I'm also interested in encoding numbers, tokenisation and NLP for low-resource languages. My background is in mathematics and previously I was a mathematics teacher in a secondary school. I also taught French and English in Colombia.



Outside my work, I enjoy salsa dancing and other latino style dances. I have also started to knit. But above all, I love travelling and get to experience different cultures.

News

  • Nov 24 - Invited Talk: Presented a high-level functioning of LLMs in Chatbot to undergraduate law students discussing risk and benefits highlighting ethical and legal considerations..

  • Jan 24 - Emergency Reviewer: Reviewed for LREC-Coling 2024, tracks "Less-Resourced/Endangered/Less-studied Languages", "Lexicon and Semantics" and "Social Media Processing".

  • Nov 23 - Huggingface Workshop: Delivered first workshop to undergraduates explaining how to train a model on custom dataset using huggingface.

  • Nov 23 - Supervision: Started jointly supervising two undergraduate students on multilingual numerical reasoning and number decoding.

  • Jul 23 - Youtube fame: Feature in AI Coffee Break youtube channel in ACL video at 8 mins 51 at ACL main conference, Toronto, Canada.

  • Jul 23 - Conference: Attended ACL 2023 in Toronto where I also presented FERMAT at the main conference.

  • Jun 23 - Quizmaster: Entertained the attendees of UKSpeech 2023 at their social event with a quiz.

  • Jun 23 - Invited talk: Presented FERMAT at the third annual Center for Doctoral Training in Speech and Language Technologies in Sheffield, UK.

  • May 23 - Field trip: Travelled to San Basilio de Palenque, Colombia to meet the afro-colombian Palenquero people and collect their native language data.

  • Feb 23 - Examiner: Marked third year undergraduate scripts for Professional Issues course.

  • Jun 22 - Conference Organiser: Student organiser of panel discussion at the Sheffield CDT second annual conferrence.

Publications

Curriculum Vitae

Find a full CV here.

Education

PhD Computer Science (Natural Language Processing)

Sep 20 - Sep 25
University of Sheffield, UK

MA Computational Linguistics - Distinction

Sep 20 - Sep 21
University of Wolverhampton, UK

PGCE Secondary Education (Mathematics)

Sep 17 - Sep 18
University of Cambridge, UK

Bachelors (MMath) Mathematics - First Class

Oct 12 - Jul 16
University of Warwick, UK
(Year abroad: Université Pierre et Marie Curie, France)

Research Activities

NLP for Endangered Language Revitalisation

Sep 21 - Present
  • Collect and digitise text data using OCR for an endangered afro-colombia language, Palenquero.
  • Deploy low-resource language NLP research to develop pedagogical resources.

Speech Technology and NLP to improve Oral History Search Functionality

Nov 21 - Jun 22
  • Project with oral history video archive to train ASR systems for transcription.
  • Adapted NER systems to identify naval domain specific terms and improve transcription search.
  • Pipeline ASR, diarisation and NER systems to provide annotated transcript used to give desired timestamps with videos.

Professional Experience

Applied Scientist (Amazon - Alexa International)

Sep 25 - Dec 25

Bellevue, Washington, USA

Trained reward models for RL using generated multilingual, multistyle cultural preference data. Conducted human preference annotation to highlight that human and foundational models preferences are uncorrelated, and depend on language and style.

Research Engineer (Huawai - Noah's Ark Lab)

Apr 25 - Sep 25

London, UK

Achieved first end-to-end autoformalisation and proof of 13 PutnamBench problems by developing decoupled conjecture-reasoning pipelines using few-shot prompting and parallel Lean4 Chain-of-Thought.

Applied Scientist (Amazon - Alexa International)

Jul 24 - Oct 24

Bellevue, Washington, USA

Optimised foundational models (Claude, Llama, Mistral) with AWS by engineering human-in-the-loop output relevancy benchmark creation process for any language. Collaborated with cross-functional teams to deploy production-ready material.

Consultant (University of Sheffield)

Jan 24 - Jun 25

Sheffield, South Yorkshire, UK

Architected and designed scalable AI solutions, including RAG for automated query response, Topic Modeling for complex document analysis, and speech-signal stress detection, while advising clients on cost-efficiency and technical stability

Teacher of Mathematics (Abbey College)

Sep 18 - Aug 21

Ramsey, Cambridgeshire, UK

Led Key Stage 5 curriculum development and university preparation, including entrance exam coaching, while mentoring trainee teachers and delivering departmental upskilling initiatives for staff.

Language Teacher (Universidad de Sucre)

Aug 17 - Jun 18

Sincelejo, Sucre, Colombia

Lectured French and English culture and language courses (A1 to C2), preparing students for official certifications and supervising student teachers on pedagogical action-research projects.

Collaborators and Affiliations

Contact

Location:

Department of Computer Science, Regent Court (DCS), 211 Portobello, Sheffield, S1 4DP, United Kingdom