Resume

Full resume text, for screen readers and applicant tracking systems:

Armaan Gulati

New York, NY · 510-514-3075 · gulatiarmaan@gmail.com · linkedin.com/in/armaan-gulati · github.com/armaangulati1

Objective

Applied-AI / forward-deployed engineer seeking to build LLM-powered products in healthcare. Biomedical engineer who designs and ships agentic systems end to end on the Claude and OpenAI APIs, from clinical data extraction to full-stack production deployment.

Skills

AI / ML:: LLM system design & application development (Claude, OpenAI & Google Gemini APIs), agentic systems & orchestration (tool-calling, multi-step router → validator → verifier pipelines), Model Context Protocol (MCP), agent identity & auth validation, RAG (retrieval-augmented generation, reranking, RAGAS), context engineering, structured/constrained generation & schema validation (Pydantic/JSON), evaluation harnesses (precision/recall/F1, LLM-as-judge, CI eval gates), prompt engineering, multimodal AI, vector embeddings & semantic search (Chroma)
Backend & Full-Stack:: FastAPI, Streamlit, REST APIs & third-party API/webhook integrations, Docker, CI/CD (GitHub Actions), testing (pytest), observability (Langfuse), Git/GitHub, Poetry, Jupyter; AI-assisted / live coding
Languages:: Python (Pandas, NumPy, Matplotlib), Java, SQL, MATLAB
Healthcare:: FHIR (resource modeling & validation, live REST API, batch/transaction bundles, Synthea, local HAPI server), clinical NLP / chart abstraction, prior-authorization / RCM workflows, PHI redaction & de-identification
Data & Tools:: Data cleaning & preprocessing, statistical analysis, data visualization, Excel (pivots), Figma

Experience

AI / Software Developer (Part-Time), Trust.care

May 2024 – Present

Architected a full-stack, AI-powered referral platform (FastAPI, Streamlit, LangChain, Chroma) connecting healthcare providers with community-based organizations (CBOs) for Medicaid Marketplace referrals; built ground-up as a two-person engineering effort, now processing 5.12M+ outreach messages to a ~955K-member Medicaid cohort across 100+ CBOs.
Designed the LLM system architecture: orchestrated Claude, OpenAI, and Google Gemini models via LangChain over a Chroma vector store, enabling multimodal (text + document/image) semantic search, plus a reply-classification pipeline that analyzed 118K+ inbound member replies by intent, separating high-signal responses from noise.
Built autonomous, tool-calling agents with an agent identity & authentication layer, giving each agent a verifiable identity to securely validate and act on people-initiated operations, addressing a core trust challenge in agent-based systems.
Drove the platform from concept to production: 19.2% cohort engagement, ~12.8x the published Medicaid SMS benchmark, contributing to 35K+ reported Medicaid enrollments and re-enrollments.

Data Analyst Intern, Sycamore Informatics

May 2024 – Aug 2024

Investigated pharmaceutical fraud by consolidating 4-5 raw financial extracts and building chunked pandas pipelines over larger-than-memory datasets, surfacing anomalous activity patterns for internal review.
Applied statistical methods and data visualization to make fraud signals interpretable for non-technical stakeholders, partnering cross-functionally to strengthen data integrity.

Technical Intern, Harmony Health

May 2023 – Aug 2023

Built predictive models (campaign response, no-show risk, enrollment likelihood) from healthcare campaign data; the no-show model was used to prioritize outreach lists, outperforming the prior targeting approach.

Projects

Clinical Prior-Authorization Agent, deployed MCP-powered agentic system

2026

Built and deployed an end-to-end agentic system automating a clinical prior-authorization workflow: authored two Model Context Protocol (MCP) servers (read + action) and an LLM agent that reads structured clinical facts from a live FHIR server (HAPI/Synthea), fuses them with free-text note extraction (with provenance), checks payer policy, and decides (submit / request-more-info / deny-risk), gated by a human-in-the-loop approval step and a deterministic guardrail.
Engineered a rigorous eval harness (93.7% macro-F1 on a held-out set, per-class precision/recall/F1, trajectory + LLM-as-judge with judge validation, CI regression gate); added PHI redaction, prompt-injection defense, retries/idempotency, and deployed the MCP server live (Docker, Fly.io, StreamableHTTP, token auth) (GitHub).

ChartExtractor, agentic clinical data-extraction platform

2026

Built an agentic pipeline (router → extractors → validator → verifier) extracting structured oncology variables from clinical notes into schema-validated JSON; 93.6% macro-F1 against a ≥85% CI deploy gate, with a per-field eval harness, containerized deploy, and a live demo + API (demo · GitHub).

Education

Washington University in St. Louis

May 2026

Bachelor of Engineering in Biomedical Engineering. Partial scholarship, College of Arts & Sciences.

WashU's undergraduate Biomedical Engineering program is ranked #6 in the nation (U.S. News 2026); the university ranks #20 among National Universities (McKelvey School of Engineering).

Relevant coursework: Optimization, Statistics, Engineering Mathematics, Differential Equations, Signal & Image Processing, Quantitative Physiology.

High school: 4.65 GPA; ranked #38 in California.