ML Researcher - NLP

About the Company

Adalat AI is building an end-to-end justice tech stack that automates manual and clerical pain points in courtrooms, giving judges back time to focus on what matters most: decision-making and delivering justice. Our solutions - from AI-powered transcription in Indian languages to case-flow management and document navigation - are now deployed across 9 states, covering nearly 20% of India’s judiciary. Backed by leading technology companies and funders, and incubated at MIT and Oxford, Adalat AI is working to eliminate judicial delays and expand access to timely justice. Founded by a team with backgrounds in law, technology, and economics from Harvard, Oxford, MIT, and IIIT Hyderabad, we are scaling rapidly across India and the Global South.

Role Overview

Indian legal language is genuinely hard. Proceedings mix formal legal English with regional languages mid-sentence. Judges dictate orders in code-switched prose. Legal documents are often scanned, hand-annotated PDFs from district courts that no OCR system handles well. There are no usable benchmarks for most of these tasks. If you want to work on NLP problems that haven't been solved yet, this is a good place to be.

In this role, you'll own the language and document understanding research agenda at Adalat. The scope runs from classical NLP problems — QnA, search, summarisation, and translation — to document intelligence and multimodal tasks that extend into OCR and vision. You'll design the experiments, define the benchmarks, curate the datasets, and get your hands dirty building models that work on problems the field hasn't caught up to yet.

This is a Research Scientist role. Your primary output is modelling judgment, experimental insight, and reproducible work. You'll work closely with engineers who take your models and make them deployable in real courtroom workflows.

Key Responsibilities

1. Language understanding and reasoning

Build and fine-tune models for legal QnA and search — systems that can answer questions over case history, statutes, and filings accurately and with citations. Develop legal reasoning and agent capabilities: models that can navigate multi-step document workflows, identify relevant precedents, and surface what a judge needs. Own multilingual summarisation and translation for legal proceedings across 10+ Indian languages.

2. Document intelligence and multimodal understanding

Develop OCR and document parsing capabilities for scanned legal filings, orders, and exhibits — the messy, handwritten, low-quality documents that existing tools fail on. Extend language understanding into vision: document layout understanding, table extraction, and multimodal reasoning over legal exhibits and evidence. Build structured information extraction pipelines from unstructured legal text and documents.

3. Evaluation methodology and benchmarking

Design task sets and benchmarks for Indian legal NLP where none exist. Define metrics that capture what matters in a legal context — not just F1, but what a judge actually needs from a summary, a QnA answer, or an extracted clause. Conduct rigorous ablations and failure analyses that produce insight, not just numbers.

4. Data curation and research dissemination

Build training corpora from courtroom documents, judgements, and proceedings across languages and court tiers. Design annotation schemes for legal NLP and vision tasks that produce consistent, high-quality supervision at scale. Maintain reproducible, publication-ready experiment logs and actively submit to venues like ACL, EMNLP, or ICLR. This is original research — treat it as such.

Qualifications

Must have

2–6 years in NLP or ML research, in industry or graduate school. Fluency in PyTorch and HuggingFace Transformers. Hands-on experience fine-tuning large language models (SFT, LoRA, RLHF, or DPO). Experience designing evaluation methodology — not just running benchmarks, but deciding what to mea