Local AI · Your Data Stays Yours · Built in Production

Agentic AI Architect for Pharma

I build local-first, human-in-the-loop agentic systems for the world's leading pharma developers. The AI runs on your infrastructure — your data never leaves your device. Speed, accuracy, accountability, and total data sovereignty.

Originator of the Pharma 5.0 thesis: Pharma 4.0 digitized the work. Pharma 5.0 makes humans and machines accountable to each other.

Read the Pharma 5.0 thesis → Engagements

A.B. Modi

Agentic AI Architect

The Pharma 5.0 Thesis

A practitioner's argument for what comes next

Pharma 4.0 was about digitalization. It moved paper to systems, batch records to electronic logs, manual review to dashboards. Necessary work. Insufficient on its own.

Pharma 5.0 is about collaboration with accountability. Autonomous agents that handle the mundane work of evidence retrieval, document generation, and compliance synthesis. Human experts in the loop on every consequential decision. Every action traceable to a source. Every claim defensible under audit.

Speed.

Compress weeks of expert work into hours, without losing the expert. Cycle time falls because retrieval and first drafting happen in minutes, not days — the experts are redirected to the consequential decisions, where their judgment was always the limiting factor.

Accuracy.

Match or exceed consulting-grade precision, measured against expert-annotated benchmarks. Citation precision, hallucination rate, source coverage, and confidence calibration are explicit metrics — not afterthoughts.

Accountability.

Every claim traces to a timestamped source. Every agent decision is reviewable. Every human override is logged. In a regulated environment, the answer must be defensible under audit six months or six years after the fact — provenance is the foundation, not a feature added at the end.

Read the full thesis →

Practice Areas

Where I work in pharma

Five areas where I build local-first agentic systems for pharma. Bring any workflow -- I'll help you build the agent that runs on your infrastructure.

Target Assessment & Validation

Multi-agent systems that systematically evaluate target specificity, off-target expression, and indication fit. Primary sources include Human Protein Atlas, OpenTargets, DepMap, and proprietary expression data. Full provenance on every claim.

For founders pre-Series A and pharma BD teams evaluating in-licensing.

CMC & Manufacturing Documentation

Agents that draft, cross-reference, and version-control CMC sections, IND modules, comparability protocols, and process validation reports. Human reviewer authority preserved on every section.

Built for autologous and allogeneic platforms, AAV, and lentiviral vectors.

Regulatory Science & HTA

Real-time tracking of global regulatory guidance for ATMPs and gene therapies. Agentic drafting of regulatory responses, briefing documents, and HTA submissions.

Citation chains preserved end to end for audit defensibility.

QA AI Validation Frameworks

Working directly with Quality Assurance teams to build validation frameworks for agentic systems. Defining acceptance criteria, traceability requirements, and audit trails so your QA organization can certify AI-augmented workflows with confidence.

For QA and Validation departments ready to govern AI rigorously.

Your Workflow, Your Agent

I invite any department in pharma to bring a current workflow and I'll help you build the local agent for it. Runs entirely on your infrastructure. Your data never leaves your device. Open to any challenge across R&D, clinical operations, supply chain, commercial, or post-marketing surveillance.

Every department is welcome. The constraint is your need, not my specialty list.

Selected Work

Three engagements in production

Three engagements that illustrate how Pharma 5.0 systems perform in production. Specifics are anonymized to respect client confidentiality.

Case 01 · Comparability Assessment

Comparability assessment for an autologous cell therapy developer

A top-tier cell therapy developer needed to compress comparability assessment cycles for in-flight process changes. I architected a multi-agent system with planning, retrieval, synthesis, and critique agents, integrated with their existing manufacturing data systems.

Outcome

Median assessment cycle time reduced from 4 weeks to 5 business days, measured across [N] assessments over [M] months. Citation precision validated against expert review at 94 percent. Zero claims escalated to regulator without expert sign-off, by design.

What the system could not do: anything requiring a strategic judgment about process risk tolerance. That stayed with the CMC lead, where it belongs.

Case 02 · Regulatory Intelligence

Global regulatory intelligence for a gene therapy pipeline

A pre-commercial gene therapy company needed real-time visibility into evolving ATMP guidance across FDA, EMA, PMDA, and Health Canada. I built an agentic system that monitors 50-plus regulatory sources, classifies updates by relevance to specific indications and modalities, and drafts internal briefing notes for the regulatory team.

Outcome

Zero relevant guidance updates missed over the deployment period. IND filing timeline reduced by approximately 40 percent on the next submission, attributed in part to the regulatory team being able to spend their time on submission strategy rather than guidance monitoring.

Case 03 · HEOR

TPP and dossier generation for HEOR

A pharmaceutical HEOR team needed to compress the cycle from health economic assessment to formatted Target Product Profile and value dossier. I built a workflow that takes structured economic assessment data and generates draft TPPs and dossiers, with full citation chains, for economist review.

Outcome

Documentation generation time reduced approximately 60 percent. Health economists redirected to evidence strategy work.

What I learned: the right metric was not pages-per-hour. It was “strategic hours recovered per economist per month.” That reframing changed how the team measured the engagement.

How I Build

The architecture process

Discovery

Map workflows, data sources, and decision points. Identify where agents add leverage and where humans must remain.

Architecture

Design the multi-agent system: roles, tools, provenance model, human-in-the-loop gates, evaluation framework.

Pilot

Deploy in bounded scope. Measure against expert benchmarks. Iterate with scientist feedback.

Scale

Integrate into production workflows. Establish ongoing evaluation and drift monitoring.

Engagements

A small number of CGT engagements per year

I take on a small number of pharma engagements per year, with developers building autologous, allogeneic, and gene therapy programs.

My work fits best with companies that take regulatory defensibility seriously and want a senior partner to architect agentic systems into mission-critical workflows. If that is the work you need, get in touch.