Nishant Keni Vol. I, 2026
Mathematics · AI engineering · OncologyOpen access

Chief Applied Scientist · Capsule AI

Nishant Keni

BTech (Electronics), MS (Electrical & Computer Engineering) MPS (Medical Oncology), MDip (Counseling Psychology) Fellowship in Clinical Oncology (AHERF), CPD-UK accredited Currently: PGD in Computer Science & Engineering, IIT Bombay

1Chief Applied Scientist, Capsule AI   2formerly Research Scientist II, Amazon  

Nishant Keni
Fig. 0. The author.
Received 1995, Mumbai Revised continuously Correspondence nishant.keni@outlook.com Status open to advisory, board & leadership conversations

Abstract

I work where mathematics, AI engineering, and oncology meet. At Capsule AI I lead applied science for agentic systems that read trials, literature, scans, and records, so that clinical and research teams can act on what they contain. Before this I spent two years as an independent AI consultant to companies in medical AI, information retrieval, and ed-tech, and nearly five years at Amazon on causal inference and ad measurement. My degrees are in electrical engineering, with postgraduate study in medical oncology.

Keywords:
Write to me Book a call LinkedIn

1Introduction

Medicine produces more evidence than any team can read. Trials, literature, imaging, and records hold answers, but in forms that resist use1. My work is building systems that read that evidence and turn it into decisions a clinical or research team can defend.

d*= argmaxa 𝔼[ U(a, Y) | do(a), 𝓔 ] (1)

Hover a term of equation (1).

Equation (1) is a compact statement of the job, and each term draws on a different discipline. The expectation and the do-operator come from mathematics. Reading and checking the evidence 𝓔 at scale is an engineering problem, and it needs evaluation and guardrails before anyone should rely on it. The utility U only means something with clinical judgment behind it2.

2Areas of work

My work falls into four areas. Each needs mathematics, AI engineering, and oncology at once; what changes is which one leads3.

Treatment planning

Models of how a tumour’s sensitive and resistant cells compete, used to design dosing schedules. Agents propose schedules, simulation tests each one, and the best are checked on virtual patients they have never seen.

Clinical evidence

Agents read trials, papers, and patient records, and extract each result with its uncertainty. Statistics pools them into one estimate, adjusted where the data are observational, that a clinical team can act on.

Evaluation & guardrails

Measuring how often agents are wrong, calibrating how sure they are, and deciding when a case should go to a clinician instead. The clinician sets the acceptable error; the mathematics finds where to draw the line.

Causal measurement

Separating what a treatment does from who happened to receive it. A comparison that looks harmful overall can reverse once the confounder is accounted for. Practised at scale at Amazon; applied now to clinical data.

Figure 1. Four areas of work, each sketched as the figure it usually ends in: tumour response under two dosing schedules; a forest plot pooling several studies; error against automation, with the clinician’s limit; and an effect that reverses once patients are compared within the same severity.

3Career

Figure 2 lays my training and appointments out on a single time axis, one row per discipline. The rows ran in parallel more often than in sequence.

Figure 2. Career across the three disciplines, 2013 to present. Bars are degrees and appointments, dots are publications and awards. Hover any mark. Tables 1 and 2 give the same data.

4Appointments

Table 1. Professional appointments, most recent first.
PeriodRoleFocus
2026–Chief Applied Scientist 1Capsule AIAgentic-pipeline architecture, harness engineering, and evaluation for AI over life-sciences data.
2024–26Principal AI ConsultantIndependentGuiding and building for companies in four areas:
  • EMR and AI clinical workflows MOC Cancer Care & Research Centre (Apr–Aug 2025)
  • AI knowledge bases i3 Digital Health
  • Silicon sampling and agentic search Keplar.io
  • Agentic automation for ed-tech CuePilot AI
Innovation Excellence award, i3 Digital Health, Q4 2025
2022–24Research Scientist II 2Amazon · Ad Measurement and OptimizationCausal model calibration, statistical noise reduction, and portfolio planning for advertising.
2019–22Data Scientist I & IIAmazon · Ad Measurement and OptimizationCausal inference and recommender systems for advertising measurement.
2018Visiting ResearcherUC BerkeleyAdversarial-ML defense with Prof. Dawn Song, on certified robustness for deep networks.

5Selected publications

  1. Adaptive containerization for microservices in distributed cloud systems. IEEE CCNC, 2020.27
  2. Neural networks based leaf identification using shape and structural decomposition. IEEE GTSP, 2016.Best Paper of Session13
  3. Content based image retrieval for leaf identification using structural features and neural networks. SPIN, 2017.6
  4. A computational intelligence approach for cancer detection using artificial immune system. Computational Intelligence: Theories, Applications and Future Directions, 2018.1
  5. Two Bayes Make Lesser Noise: A Risk-Aware Blind Filter for Improving Measurement Precision. Amazon Consumer Science Summit (CSS), 2022.confidential

Bars show citations to date. Full list on Google Scholar →

6Education

Table 2. Degrees and fellowships, most recent first.
YearQualificationInstitution
2025–PG Diploma in progressComputer Science & EngineeringIIT Bombay
2024–25FellowshipClinical Oncology · CPD-UK accreditedApollo Hospitals Educational and Research Foundation (AHERF)
2024MDip4Counseling PsychologyIISDT
2023–25MPSMedical OncologyTech Global University
2019–21Professional CertificateArtificial IntelligenceStanford University
2017–19MSElectrical & Computer EngineeringGeorgia TechGPA 4.0/4.0
2013–17BTechElectronics EngineeringVJTI, MumbaiGold Medal, Department Rank 1

Also certified in ACLS (Advanced Cardiovascular Life Support) and BLS (Basic Life Support), American Heart Association.

Correspondence

If you are working at the intersection of AI and medicine, I would like to hear from you.

Open to advisory, board, and leadership conversations.

nishant.keni@outlook.com Book a call

Read in 47 countries and regions

3,730 readers since January 2023

and 42 more