Home / Company
Curator Research
Indian deep-tech for evidence-bound knowledge work.
Curator Research builds evidence-bound AI and knowledge infrastructure for Indian regulatory and document-heavy work: software, models, graphs, document intelligence and governed workflows designed around privacy, cost, evaluation and human authority.
—
The company
We build evidence-bound AI and knowledge infrastructure for Indian regulatory and document-heavy work. Our software connects source documents to structured knowledge through models, graphs and document intelligence, then supports research, change analysis and governed workflows. Privacy, cost, evaluation and human authority shape how that infrastructure is designed.
—
A regulatory knowledge ambition
Our knowledge-base design brings together regulatory directions, rules, regulations, circulars, orders and related source material from eleven Indian regulatory and government bodies: RBI, SEBI, IRDAI, PFRDA, IBBI, NABARD, IFSCA, MCA, CBDT, CBIC and TRAI. It also connects historical news and events with time-aware financial and market context.
—
Our point of view
A useful AI system should identify the right source, preserve lineage and time, explain its basis, show uncertainty and keep action inside an approved boundary. Precision, provenance, cost governance, robustness and ongoing evaluation are product requirements, not afterthoughts.
—
From source to action
We connect document processing, structured knowledge, retrieval, citations, historical context, impact analysis and supervised execution. Each engagement starts with the source material, access boundary and decision to support. The record remains inspectable from input to outcome.
How the work developed
The systems behind our products grew through successive research and engineering milestones.
- October 2025 — market-data infrastructure, then the first regulatory knowledge work
- From November 2025 — source collection from regulators and tribunals, later run as a self-scheduling fleet
- February to April 2026 — Version 1 ingestion and retrieval pipeline development, with graph-RAG research into deep contextual, semantic and source-lineage connections
- May to July 2026 — a bounded period of multi-GPU H100 access for document parsing and enrichment at scale
- July 2026 — fine-tuned models and a trained reranker migrated for use in further research and the knowledge base
- 24 August 2026 — Curator Research Private Limited incorporated
- September 2026 — Patra, BanyanGraph, Curator, ChangeProof and Karta, each in development
That work became a company aimed at the difficult part of applying AI to regulated environments: source quality, regulatory structure, model evaluation, retrieval, graphs, infrastructure, evidence and safe operation.
—
Who runs the company
An Indian deep-tech company with an all-women board. Curator Research was founded and is directed by Bhavika Mukherjee and Aditi Chadha. We plan to expand in phases, bringing in technical and functional specialists, collaborators and team members as products and engagements mature, while using automation to keep the operating model cost-conscious. The company is self-funded and has taken no outside investment.
—
Company record
Curator Research Private Limited was incorporated on 24 August 2026 in Panchkula, Haryana.
- MCA incorporation — registered under the Companies Act, 2013; CIN U62011HR2026PTC149626
- MSME registration — Udyam Registration Number UDYAM-HR-13-0040757; classified as a Micro enterprise for 2026–27 as of 19 September 2026
- Company PAN — AAOCC1817G
- Company TAN — RTKC10932F
—
Where the work runs
We design systems around the environment that owns the data. Patra's document-processing core is built around verifiable no-egress execution. The private knowledge and retrieval path is under development, with local and air-gapped operation as design targets. Where licensing and evaluation permit, we prefer open-weight, self-hosted models for control over data, cost and availability. Where a client requires frontier-model capability, approved models can be integrated within the agreed tenancy and data-handling architecture, with privacy-protected or restricted content kept behind explicit boundaries and only the minimum permitted context leaving the client-controlled environment.
- India-built company, with the initial knowledge context centred on Indian regulatory material
- Verifiable no-egress document processing today; local/private knowledge operation is a direction under development
- Model choices follow licensing, evaluation, client approval and privacy requirements; any frontier model receives only the minimum permitted context
—
Working with what you already have
We scope each engagement around the organisation's sources, permissions and decisions. We test access boundaries before sensitive data is admitted, then connect the approved material into usable, cited context.
—
How we operate
We run the company on an operating layer we are building for ourselves. It keeps authority, decisions, project state and approvals explicit, so a small team can work with many AI agents without losing control. It is an internal system, not a product.
Bring us the difficult part of the work.
Tell us about the documents, the decision and the constraints. We start with scope and evidence, propose a solution, and then decide together what kind of engagement is useful.