Senior Data Engineer Mumbai · 4+ years · Full-time →

NyaayOS; Mumbai; On-site; INR 3,000,000 - 5,500,000; Full time

Job summary

NyaayOS is hiring for a backend data-engineer role focused on building scalable ingestion pipelines for large volumes of unstructured Indian legal data. You will tackle unreliable sources, ensure correctness, and drive reliability across the platform.

Ideal candidates bring 4+ years of backend experience, strong Python, AWS proficiency, and hands-on OCR expertise. Expect to implement idempotent pipelines and robust observability in a fast-moving environment.

Qualifications

• 4+ years building data-heavy backend systems.
• Experience with production data pipelines and reliability.
• Hands-on OCR and document processing tools.

Responsibilities

• Own ingestion pipelines across large volumes of unstructured documents.
• Ensure idempotency, retries, partial-failure recovery, and observability.
• Develop asynchronous workflows (Airflow, Celery, Dagster, Temporal, or similar).
• Build production-ready, observable data code and pipelines.

Skills

Backend systems

Data pipelines

Python

AWS

Observability

Education

B.E./B.Tech in CS or a related field

Tools

Tesseract

Airflow

Celery

Job description

Every answer ParamAIQ gives must trace back to an exact document, paragraph, and source. That guarantee starts with the pipeline you build. You turn messy legal records into clean, searchable, reliable law: scanned PDFs, handwritten orders, government portals that change without warning. This is the foundation the whole product stands on, and it is not scraping. It is the difference between an AI answer we can defend and one we can't.

YOU WILL OWN

• The ingestion platform that collects, processes, and structures Indian legal data at national scale, across courts, tribunals, statutes, and gazettes.
• Document understanding for records of wildly varying quality, from clean digital filings to low-quality regional-language scans.
• The classification and data-quality gates that decide what is good enough to serve to a system that must be citable.
• Provenance across the pipeline, so every derived artifact traces back to its source.
• Safe reprocessing, so improved models can re-run over the corpus without disturbing live systems.

THE HARD PROBLEMS

• Staying correct across hundreds of unreliable sources. Portals go down, change format overnight, and lie in their metadata. Ingestion that survives all of it and never silently drops a record is the core reliability problem.
• Knowing how much to trust a document. Quality ranges from machine-readable to barely legible handwriting, so the system must attach a confidence to what it extracts, not just extract it.
• Reprocessing at scale without breaking trust. When a model improves, you rebuild parts of the corpus reproducibly and reversibly, while the product keeps serving.

WE ARE LOOKING FOR

• 4+ years building data-heavy backend systems, including production pipelines over large volumes of unstructured documents.
• A real grasp of pipeline reliability: idempotency, retries, partial-failure recovery, observability.
• Hands-on OCR and document processing (Tesseract, Textract, layout-parser, Donut, or similar).
• Async pipeline experience (Celery, Airflow, Dagster, Temporal, or similar).
• Strong Python and clean, production-ready data code.
• Working AWS: S3, IAM, RDS, Lambda or Fargate, large-scale object storage.
• You handle broken PDFs, bad encodings, missing metadata, and shifting source portals without flinching, and you ship reliable systems without over-engineering them.

NICE TO HAVE

• Running scraper and collection platforms at scale: rate limits, proxy rotation, schema drift, source failures.
• Multilingual or Indic NLP: Hindi, Tamil, Marathi, others.
• Document AI, legal-tech, e-discovery, or other document-heavy pipelines.
• Indian legal sources: eCourts, e-SCR, IndiaCode, e-Gazette.
• Knowledge graphs, citation graphs, entity resolution.
• DPDPA and data-residency awareness

THIS ISN'T THE ROLE IF

• You want clean data and clean specs. Indian legal data is genuinely bad, and most of the job is a system that stays correct anyway.