Chandra Vikram · Healthcare AI/ML Engineer

Experience

2025 – nowHealthcare AI/ML Engineer @ Takeoff41, a Stanford spin-outEnd-to-end ownership of TPN2.0, AI clinical decision support for neonatal nutrition: product and application development (frontend and backend), system architecture, Azure cloud infrastructure, CI/CD, and release management; LLM fine-tuning (LoRA/QLoRA) with synthetic data and SFT pipelines; LLM evaluation, safety, and adversarial testing; hybrid retrieval (RAG); model deployment and monitoring on Azure; EHR integration; MCP servers; agentic engineering with AI coding agents (Claude Code, Codex, Pi)
2024 – 2025Health Technology Developer @ IU Health, MS capstoneSMART on FHIR patient app; EHR integration; REDCap patient-reported outcomes; LLM assistant for provider summaries
2024System Developer Intern @ Caring Community PartnershipDHIS2 deployment; healthcare data warehouse; ETL pipelines; predictive models for mental-health programs
2023 – 2025M.S. Health Informatics @ Indiana UniversityClinical Decision Support Systems; Machine Learning for Health; Health Interoperability Standards and Terminologies; Clinical Information Systems; Applied Statistical Methods for Biomedical Informatics; Natural Language Processing for Biomedical Records; EHR System Design
2018 – 2022B.S. Pharmaceutical Sciences @ RGUHS, BengaluruPharmacology, Pharmaceutical Engineering, and Pharmacovigilance

Hi, I'm Chandra Vikram. I work to bring AI into clinical care in ways clinicians and patients can trust. Currently, I'm a Healthcare AI/ML Engineer at Takeoff41, a Stanford spin-out, where I work on TPN2.0, AI clinical decision support for neonatal parenteral nutrition, now in use at multiple hospitals. The research behind it was published in Nature Medicine in 2025.

At Takeoff41, I own the engineering end to end: designing the system architecture, building the application from frontend to backend, running our Azure cloud infrastructure, CI/CD, and releases, fine-tuning open-weight models on clinical data, building retrieval and evaluation pipelines, deploying and monitoring models in production, and integrating the system with electronic health records so its recommendations appear inside clinicians' existing ordering workflow. I built the evaluation pipeline for our study on large language models in parenteral nutrition prescribing, accepted at JAMA.

Previously, I built Carepal with IU Health, a SMART on FHIR app for liver cirrhosis management that was a finalist in the 2024 AMIA/HL7 FHIR App Competition in San Francisco. Before that, I deployed DHIS2, the open-source health information platform used by national health programs worldwide, and built data pipelines on it for a community health nonprofit in Missouri.

I came to software and AI from pharmaceutical sciences, which I studied in Bengaluru, before my Master's in health informatics at Indiana University. What kept pulling me back was watching, in hospitals and clinics, how often a good clinical decision still depended on whether the right piece of patient information reached the right clinician at the right time, and how rarely healthcare systems made that easy. Closing that gap is what drew me to healthcare AI.

I believe healthcare AI, done with rigour and with clinicians firmly in the loop, is among the most transformative technologies of our generation, and that the wrong systems can do real harm. Those of us who build it share a responsibility to keep it safe, accessible, and able to scale, so that it serves patients, the people who care for them, and humanity more broadly.

Outside work, I play guitar and football, and I'm usually reading about history and ancient Indian philosophy, or deep in a documentary.

Below are my working notes on clinical AI: evaluation, retrieval, safety, and FHIR tooling. Always happy to talk shop, so feel free to reach out.

Research & Projects

Projects I'm actively building and researching across clinical benchmarks, medical LLM safety and alignment, voice and telehealth agents, EHR-navigating agents, clinical knowledge graphs, retrieval evaluation, and FHIR infrastructure. Each is at a different stage, from early design to working prototype, and each write-up covers the problem, the approach, and the research it builds on.

01	Triagemind	Triage	A four-agent ED triage system with calibrated uncertainty, red-flag screens, and structured handoff, pinned to Sax et al.'s 32.2% ESI mistriage baseline.
02	Graphcore	GraphRAG	Microsoft GraphRAG methodology applied to clinical guideline corpora, the first cost-quality Pareto curve for clinical GraphRAG.
03	Atrium	FHIR	A reference Model Context Protocol server exposing FHIR clinical data via SMART-on-FHIR launch.
04	Reason·Med	Fine-tune	An open clinical reasoning model trained via continued pretraining, SFT, and GRPO on Qwen3-8B.
05	Chaincite	Citations	Clinical RAG benchmark measuring citation correctness AND faithfulness, built on Wallat et al.'s ICTIR 2025 distinction.
06	Pharos	Chronic	Long-horizon voice agent for HF and type-2 diabetes with persistent memory and clinician oversight loop. Targets 80% adherence at 12 weeks.
07	Longitude	Long-context	Diagnostic reasoning over decade-long longitudinal patient records, 150k–500k tokens each.
08	Oracle	Diagnosis	A differential-diagnosis agent that emits evidence-grounded reasoning traces, citation per claim.
09	Auris	Scribe	Ambient clinician-patient voice to validated FHIR resources, end-to-end.
10	Medigraph	Knowledge-graph	Patient-centric clinical knowledge graph from FHIR + clinical notes + UMLS / SNOMED / RxNorm / LOINC. Composes with Atrium.
11	Asclepius	Red-team	An adversarial benchmark for medical LLM jailbreaks and sycophantic capitulation.
12	Calline	Voice	Voice-first after-hours nurse triage agent: streaming ASR, gpt-realtime, sub-second TTS, uncertainty-gated escalation.
13	Caliper	Benchmark	A FHIR-grounded extension of HealthBench with a public cross-model leaderboard.
14	Conscience	Alignment	Constitutional AI applied to clinical decision support, fine-tuned on Qwen3-8B.
15	Telesight	Telehealth	A three-phase telehealth copilot covering pre-visit chart prep, intra-visit CDS with Five-Rights gating, and post-visit instructions plus coding.
16	Chartwalker	Computer-use	A Claude-driven agent navigating a real EHR interface with a deterministic grading harness.
17	Ragprobe	Adversarial	Adversarial robustness benchmark for clinical RAG: PoisonedRAG / BadRAG / GARAG / Phantom / indirect-injection on clinical corpora.
18	Vestibule	Transitions	Post-discharge transition agent with a 24h/48h/72h/7d voice-call cadence pinned against Jencks's 19.6% 30-day Medicare readmission baseline.