AI ENGINEER / APPLIED NLP

Rudra Narayan Meher

Building reliable AI systems for complex documents and underserved languages.

I build evidence-grounded AI systems, from medical-legal document intelligence to low-resource language translation. Co-founder of Shivru AI and a Data Science student at IIT Madras.

Explore selected work Download résumé

IIT MadrasBS Data Science & Applications · 2024–2028 (expected)

Selected work

All projects

01 / Document intelligence · Development MVP

Shivru AI

Turn fragmented medical-legal records into reviewable timelines, grounded answers, and source-linked reports.

500 synthetic pages
251 automated checks
50 browser tests

Documented 11 Oct 2026 · synthetic load and software checks, not clinical accuracy.

Read the engineering case study →

Six-stage Shivru pipeline from private PDF upload through OCR, typed extraction, grounding, retrieval and human review. — Simplified architecture from implementation documentation.

Conceptual retrieval interface: a legal source document passes through OCR, hierarchical JSON and chunking to a Qdrant source excerpt. Not an application screenshot.

Prototype

Advotac

Indian legal research with OCR, structured legal documents, and Qdrant retrieval.

QdrantOCRAWS

Explore project

Conceptual time series showing a historical trend with seasonal variation and a dashed hypothetical forecast region. No measured prices or accuracy are shown.

Prototype

Mandi Price Forecasting

Agricultural price forecasting with historical mandi data and seasonal patterns.

Machine learningTime series

Explore project

Language, data & research

Research overview

Low-resource language research

SamLingo — A Language Model for Sambalpuri

V2 refines the data and continues training for English-to-Sambalpuri translation in Odia script.

V1 → V2 · reported evaluation: 6.78 → 33.54BLEU
V2 update: 6,400 records
Final benchmark: 149 reference pairs

Author-corrected reported comparison; test-set independence is not verified.

View SamLingo project Evaluation & evidence Hugging Face model

A little background

My work sits between applied machine learning, document intelligence, and multilingual NLP. In Shivru AI, a date needs a role: a print date must not become the date of a clinical event. I keep findings tied to the original page and excerpt so reviewers can check that distinction.

More about my approach →

Let’s collaborate

Building something in applied AI, NLP, or document intelligence?

Open to research collaborations, engineering opportunities, and interesting technical problems.

rudra.meher@rudrameher.in