Maitreya Sapariya

I build software systems that report what they couldn’t resolve.

I’m a software engineer with a degree in AI and data science, so I work on both halves: the pipelines, and the models that run on them. My research is about the decisions inside those systems, at design time, before any labels exist, and at runtime, when the data can’t settle the answer. Two of my papers come at this through accessibility and AI ethics: who receives an answer, in what form, and whether they are told when the model is unsure.

Software Engineer, Crest Infosystems · Surat, India

Maitreya Sapariya standing outdoors in a black blazer and glasses, smiling

Publications and preprints

published

When Chats Split Apart: Formalizing Temporal Communication Fragmentation and the CARTA Reconstruction Framework

Maitreya Sapariya. International Journal of Management and Data Analytics (IJMADA), Vol. 6, Issue 1, 2026, pp. 449–470

presented

What AI should refuse to say, and to whom

Maitreya Sapariya. Workshop on Human-Centered Perspectives on the Role of AI in Accessibility, Academic Mindtrek 2026, Tampere (hybrid)

submitted

Three Voices: Uncertainty-Gated Rendering of a Maternal-Health Risk Alarm for Mothers, Community Health Workers, and Clinicians

Maitreya Sapariya, Aditi Patil. Submitted to arXiv, awaiting announcement; code and artifacts archived on Zenodo

completed

Prediction is not selection: a pre-registered study of design-time mechanism choice in author name disambiguation

Maitreya Sapariya. Independent research; pre-registered study complete, not submitted to a venue

Experience

Jan 2026 – now

Software Engineer · Crest Infosystems, Surat

Full-time since July 2026, converted from a 6-month internship

Software Engineer responsible for building scalable systems that people actually use: thinking through, integrating, and building AI solutions for existing systems.

Jun – Jul 2025

Data Analyst Intern · InnoByte Services

Internship, June to July 2025. Exploratory analysis of an Amazon India sales report, 128,976 line items across 120,229 orders.

Projects

all projects

in production

eRESI: AI knowledge assistant for a mortgage lender

client work · Crest Infosystems

A production Q&A assistant over a mortgage lender's guideline PDFs. Every answer links to the sentence it came from, and questions it can't answer go to a human underwriter.

archived