Research as a public practice.
Working themes, questions, and notes toward rigorous applied AI — the long path of a scientist-builder.
This center treats research as continuous: clear questions, careful systems, and notes that others can follow. Ambition is simple and demanding — contribute at a world-class standard through applied intelligence, not slogans.
Where intelligence meets engineering
AI Systems
Designing and deploying systems that reason, adapt, and integrate into real workflows.
Automation Thinking
Mapping decisions into pipelines — less repetition, more signal.
Intelligent Workflows
Orchestration across models, data, and human intent.
Scalable Architectures
Modular, observable, production-ready backends.
Standards over noise
Questions before claims
Publish the problem clearly. Results follow method.
Systems over demos
Prefer architectures that survive contact with reality.
Open notes
Working notes and articles document the path — see Writing.
Current project
AI Doubt-Solving Assistant for Hindi and English Learners
Many Indian learners think and ask in Hindi, but most AI tools are tested mainly in English. This project studies how well large language models answer real student doubts in both languages — and whether grounding answers in course notes makes them more accurate.
Research questions
Language gap
How does LLM answer quality differ when the same student doubt is asked in Hindi versus English?
Grounding
Does retrieving from course notes (RAG) reduce wrong or made-up answers compared with a plain LLM?
Learner trust
Do students find grounded answers clearer and more trustworthy than ungrounded ones?
Method (planned)
Collect
50 bilingual student doubts across Python, data, ML and GenAI.
Answer
Generate answers with and without retrieval, in both languages.
Evaluate
Score correctness, clarity and language quality with a simple rubric.
Publish
Release data openly and write up findings for a peer-reviewed venue.
Research log
- Sep 2026Registered ORCID iD 0009-0003-8679-1981 — public research identity.
- Sep 2026Defined project scope and three research questions.
- Sep 2026First version of the 50-question bilingual dataset (English + Hindi).
- NextBuild baseline answers and evaluation rubric.
Publications
In preparation
First paper planned from this project. Listed here and on ORCID once submitted.