The words both rooms use.

A large neural network trained on broad data that can be adapted to many specific tasks. GPT-class language models, Claude, Gemini, and Llama are foundation models; so are the multimodal and protein-folding variants.

In IndiaIndian pharma R&D and regulatory teams overwhelmingly evaluate hosted foundation models from Anthropic, OpenAI, and Google before considering open-weights deployment, primarily because of data-residency and procurement comfort with named US vendors.

A pattern where a language model answers a question by first retrieving relevant documents from a private corpus and then composing the answer from them, rather than relying only on what was in its training data.

In IndiaThe first AI workload most Indian pharma companies put into production: RAG over the company's own regulatory submissions, SOPs, and internal study reports, behind the firewall, queryable by non-technical reviewers.

Continuing the training of a foundation model on a smaller, task-specific dataset so it learns the patterns of that domain. Distinct from prompting (which uses the model as-is) and from RAG (which adds documents at query time).

In IndiaLess common in Indian pharma than RAG today: the data governance work needed to assemble a fine-tuning set responsibly often exceeds the value of fine-tuning over a strong base model with retrieval, and most teams reach the latter first.

The engineering discipline around productionising machine-learning systems: data pipelines, model versioning, evaluation, deployment, monitoring, and rollback. The pharma equivalent of GxP discipline applied to ML systems.

In IndiaOften the gating constraint on Indian pharma AI projects moving from proof-of-concept to production: a working model demonstrated on a laptop is two thirds of the way through the engineering work, not nine tenths.

Language-model systems that take actions in a software environment over multiple steps: reading a document, calling a tool, writing to a database, asking the user a follow-up. Distinct from a single-prompt chatbot in that they have a loop, tools, and state.

In IndiaThe category of system currently being piloted for regulatory submission drafting, pharmacovigilance case triage, and clinical-trial document review at several Indian pharma companies in 2025 and 2026.

The systematic measurement of an AI system's performance against a defined set of tasks and acceptance criteria, run repeatedly as the system changes. For language-model systems, evals are typically a mix of automated grading and human review.

In IndiaThe most under-invested layer of Indian pharma AI deployments today: a model that is good enough to demo but does not have eval discipline behind it cannot defensibly be put in front of a regulator or a patient-facing workflow.

A US FDA submission for approval to market a generic version of an already-approved drug. Distinct from a New Drug Application (NDA) in that it does not require independent clinical trials; instead the applicant demonstrates the generic is bioequivalent to the reference product.

In IndiaIndia is the world's largest source of ANDA filings to the US FDA; the document-heavy, response-heavy nature of the process is one of the most concrete and immediate places generative AI is being applied inside Indian generics companies.

The study type used to prove that a generic drug delivers the active ingredient to the bloodstream at the same rate and extent as the reference product. The core scientific evidence behind an ANDA.

In IndiaBA/BE studies are routinely conducted at Indian clinical research organisations and contract manufacturers; the document streams these studies produce are a natural fit for retrieval-augmented review.

A company that provides pharmaceutical development and manufacturing services to other pharma companies, typically including formulation, process development, scale-up, and commercial manufacturing.

In IndiaIndia has a large and growing CDMO sector serving global and domestic clients; CDMOs are often among the more aggressive adopters of AI for manufacturing yield prediction, deviation investigation, and batch-release documentation.

India's national drug regulator, the equivalent of the US FDA. Operates under the Ministry of Health and Family Welfare and is responsible for approval of new drugs, clinical trials, and import licences.

In IndiaCDSCO has begun issuing guidance on the use of digital tools in clinical research and pharmacovigilance; how it will treat AI-generated regulatory content is one of the open questions the consortium expects to discuss at edition.

The science and activities relating to the detection, assessment, understanding, and prevention of adverse effects of medicines once they are on the market. A regulated function with mandatory reporting timelines.

In IndiaIndian pharma pharmacovigilance teams handle large volumes of inbound case narratives in multiple languages; foundation-model-assisted case triage and translation is one of the highest-traffic AI use cases in the industry today.

The schedule of India's Drugs and Cosmetics Rules that lays out Good Manufacturing Practices (GMP) for pharmaceutical manufacturing facilities. Revised Schedule M, effective in phases from 2024, brings Indian standards closer to WHO-GMP.

In IndiaCompliance with revised Schedule M is one of the most active operational concerns at small and mid-sized Indian pharma manufacturers in 2025 and 2026; AI for documentation, batch-record review, and audit-trail analysis is being evaluated against this deadline.

The standard way Indian pharma companies organise what they sell. A therapy area (or 'cluster') contains several divisions; each division sells a portfolio of brands; each brand ships as one or more stock-keeping units (SKUs) by pack size, strength, or formulation.

In IndiaMost AI roadmaps inside Indian pharma confuse the layers: a model that predicts at the SKU level is useless to a Brand Manager planning quarterly campaigns, and a model that aggregates to division-level is useless to a regulatory team filing variations on a specific SKU. Naming the layer is the first conversation.

The reporting pyramid of the in-person sales organisation. National Sales Manager at the top; under them, Regional Sales Managers (RSM) covering large multi-state regions; under each RSM, Area Sales Managers (ASM); under each ASM, Territory Managers (TM); under each TM, the Medical Representatives (MR) who actually visit doctors and chemists.

In IndiaThe MR is the only person in the field who actually meets the prescriber. Any AI workload aimed at commercial productivity that does not have a clear answer to 'how does this change what the MR does on Monday morning' is targeting the wrong layer of the pyramid.

An HQ-side marketing role that owns one or more brands across the country: positioning, campaign strategy, sales targets, KOL engagement, promotional materials, launch plans. The marketing counterpart to the field-force pyramid.

In IndiaBrand Managers are the most frequent operational consumers of AI for Indian pharma commercial functions. The questions they ask of AI are concrete (which territories are under-indexing on Brand X relative to IQVIA market share, which MRs are over-detailing low-Rx product, which doctors are switchable) and they hold the budget that turns AI pilots into production deployments.

How Indian pharma slices the country for sales planning, target-setting, and attribution. A territory (often called an HQ, short for headquarter) is the unit a single MR covers; territories aggregate up through region/state, zone, and ultimately all-India.

In IndiaAlmost every interesting commercial AI question in Indian pharma is a territory-level attribution problem in disguise: did secondary-sales movement in HQ X come from the MR call plan, from a Brand Manager promotion, from a chemist incentive, or from a market shift the IQVIA layer would have detected?

Primary sales are what the manufacturer ships to stockists; this is what the company books as revenue. Secondary sales are what stockists ship to retailers (chemists); this is what is actually moving into the market and, eventually, to patients. The two numbers can diverge significantly in any given month.

In IndiaThe gap between primary and secondary is one of the most-misunderstood concepts by AI vendors selling into Indian pharma. A demand-forecasting model trained on primary-sales data will mis-predict shelf availability; a sales-attribution model that ignores secondary will reward MRs for moving stock into a stockist rather than moving product to patients.

The dominant third-party market-intelligence dataset Indian pharma companies subscribe to. Reports the brand's market share (MS), its rank within its therapeutic class, and the absolute market value of that class, sliced by geography and by SKU. Published monthly with a lag.

In IndiaAlmost every Brand Manager review in Indian pharma opens with the latest IQVIA numbers, and almost every AI use case that touches the commercial function has to reconcile against them. A workload that improves an internal metric while moving IQVIA in the wrong direction is dead on arrival, regardless of how technically interesting it is.

The standard Indian prescription-audit dataset. Audits a panel of doctors to estimate, for each brand, how many prescriptions are being written in each therapy area and geography. Distinct from IQVIA market-share data, which tracks what moved through chemists, not what was prescribed.

In IndiaSMSRC Rx and IQVIA secondary are the two halves of the commercial truth in Indian pharma: one says what doctors prescribed, the other says what patients (or chemists) actually moved. The interesting AI work begins where the two disagree, and the team that can explain the disagreement usefully is the team worth listening to.