I build production AI and the research behind it.

Hi! I'm Achal Dixit. I design, ship, and research production AI across pharmaceuticals, clinical trials, logistics, and genomics: high-stakes, highly regulated environments where a model has to be right, explainable, and auditable. I take systems from business needs to shipped product, and publish the research behind them.

See the work → Read the research LinkedIn ↗
$15M+
Impact delivered · reported
10+
AI systems in production
6+
Years in applied AI & research
4
Papers & talks published
01 Selected Work

Systems I've shipped

Production AI in regulated, high-velocity environments. Full case study: problem, approach, architecture, and result.

Indication φ: an engine for drug repurposing
AI FOR SCIENCE · PRODUCT SPIN-OUT
An drug repurposing engine, built end-to-end, that reads a directional, time-resolved map of drug↔disease interactions from population-scale patient timelines(1.2B+), fused with mechanism and genetics into an explainable evidence graph. A full production platform; validated by rediscovering established biology before trusting it on novel bets worths billions into trials.
Representation learning · knowledge graphs · shipped v1, end-to-end →
GenAI automation for Clinical Study Reports
BRISTOL MYERS SQUIBB
Automated Tables, Figures & Listings(TFL) generation over a clean analytics layer, retrieving from structured truth for Clinical study reporting, supporting medical writers across biologics and small-molecule programs.
$4.5M external vendor spend saved · ~2× writer productivity (reported) →
Inspection & regulatory-readiness intelligence
BRISTOL MYERS SQUIBB
LLM-driven ingestion of FDA 483s, internal audits, and deviations to surface systemic risk and remediation roadmaps for global supply leadership. Presented at BioIT World, Boston 2025.
~50% audit-prep reduction · Science & Innovation Award 2025 →
Variance Diagnostics Copilot
DELHIVERY
An AI that narrates why KPIs move, fusing causal inference, temporal clustering, and natural-language explanation for operators across 1,000+ last-mile centers.
Reporting cycle: one week → minutes →
Malaria-vector genomics ML
IMPERIAL COLLEGE LONDON
Reproducible ML workflows over ~4.8M genomic features for Target Malaria: automated QC, LD pruning, and rare-allele detection in the Burt Lab.
Genetic-origin classification · interpretable ML →
vibebill: what did this commit cost?
OPEN SOURCE · SHIPPED
Git-native cost accounting for AI-assisted repositories. It reads your coding agents' local logs (Claude Code, Codex, Gemini, aider), joins them to git history, and reports what every commit cost: measured, never guessed. It metered its own build, commit by commit.
Shipped on npm · MIT · TypeScript · exact bigint pricing →
View all projects →
02 Experience

Where I've delivered

2026 — Now
Sanofi
AI Product & Strategy Lead
Leading AI product strategy for Digital Clinical Development in R&D: building AI products, defining roadmaps, shaping governance, and driving enterprise-scale adoption.
2024 — 2025
Bristol Myers Squibb
Lead Data Scientist
Led AI across inspection readiness, regulatory knowledge discovery, clinical-trial optimization, and global manufacturing intelligence. Mentored 10+ data scientists. Science & Innovation Award, 2025.
Ongoing
Stealth AI startup
Founding Data & AI Scientist · Fractional CTO
I own the design and end-to-end build of core AI systems: a >5TB medallion-architecture lakehouse on Databricks, large-scale pipelines across 6 countries, LLM orchestration, backend services, and deployment. Built the "DataFlywheel" synthetic-data engine for privacy-preserving healthcare datasets, and automated LLM content pipelines. I translate business goals into technical roadmaps and ship production systems, not prototypes. Indication φ is a product spin-out of this work.
2022 — 2024
Delhivery
Data Scientist
Built and deployed core ML for one of India's largest logistics platforms: return-to-origin, delivery-success, and closure models running nationwide, plus a causal variance-diagnostics copilot.
2022
ZS Associates
Business Technology Solutions Analyst
Big-data analytics for global pharmaceutical clients: HCP360 data lakes and analytics pipelines across on-prem clusters, cloud, and commercial data.
2020 — 2021
Imperial College London
Visiting Research Scholar
Genomics intelligence for Target Malaria: genetic-origin classification from SNPs using interpretable ML, Burt Lab, Department of Life Sciences.

03 Education

The University of Texas at Austin
MS, Data Science & Artificial Intelligence
Pursued concurrently with industry roles at Sanofi and BMS.
Imperial College London
Visiting Research Scholar · Dept. of Life Sciences
Genomics + ML for Target Malaria: genetic-origin classification of malaria vectors using SNPs and interpretable ML in the Burt Lab.
IIIT Guwahati
B.Tech, Computer Science & Engineering
Specialized in ML, optimization, and secure systems. Research accolades in AI for healthcare.
04 Advisory

Working together

Alongside my full-time work, I take on a small number of advisory and build engagements where deep applied-AI experience in regulated, high-stakes settings moves the needle.

01
AI strategy & architecture
From capability maturity to a shippable roadmap: what to build, what to buy, and how to make it safe and auditable. Fractional CTO/CDO capacity for life sciences and beyond.
RoadmapsGovernanceAI risk
↗
02
GenAI & agentic systems, built right
LLM-native applications with proper retrieval, evaluation harnesses, guardrails, and audit trails from day one, not a demo that breaks in production.
RAGLLM evalAgentic workflowsGxP guardrails
↗
03
Regulatory, clinical & scientific AI
Validated analytics for FDA-readiness, trial optimization, pharmacovigilance, and drug-intelligence pipelines, with explainability that survives an audit.
Clinical AIDrug intelligence21 CFR Part 11
↗
05 Research

Two papers on trustworthy AI

Currently in submission. Rigorous methods for evaluating and safely deploying AI in healthcare: the science behind the systems I ship.

Indistinguishable Is Not Trustworthy
IN SUBMISSION
What: a benchmark for synthetic patient data. Finding: data judged "realistic" by every standard metric can still distort the causal treatment-effect estimates it's used for, and the generator with mathematically perfect statistics is the worst for real analysis.
5–20×hidden causal error, quantified
3national cohorts
Openreleased benchmark
Causal evaluation · synthetic data →
Human-Completing AI
IN SUBMISSION
What: a compliance-native retrieval system that drafts pharmaceutical regulatory responses under hard constraints: grounded generation, mandatory human review, enforced data boundaries. Result: validated in a 12-week pilot.
78%retrieval precision@5
0boundary violations / ~9,500 ops
46%faster first draft
Trustworthy RAG · guardrails →

Plus peer-reviewed clinical ML: American Journal of Emergency Medicine and Computing in Cardiology, 2021. See all research →

06 Recognition

Science & Innovation Award
Bristol Myers Squibb · 2025
Speaker, BioIT World Conference
Boston · 2025
MIT COVID-19 Challenge — Winner
5,000+ global participants
MHRD Samadhan Challenge — Winner
800+ teams · Govt. of India
Microsoft Imagine Cup — India Finalist
Top teams nationally
07 On air

Machine learning in healthcare

A conversation on machine learning in healthcare and getting research into production.

StatQuest with Josh Starmer
Featuring Achal Dixit · ML in Healthcare
Watch on YouTube ↗

Let's build
something real.

Open to the right role, research collaborations, and a small number of advisory engagements. If the problem is hard and the stakes are real, I'm interested.

Emailachaldixit22@gmail.com
Send a message LinkedIn ↗ GitHub ↗