Sofia Hirschmann

Founding Engineer, Elevare Health

hirschmann.so@northeastern.edu · linkedin.com/in/sofiadhirschmann · github.com/sofiahirschmann · sofiahirschmann.com

AI engineer building clinical AI for maternal care. I lead engineering at Elevare Health, where I architect the AI agent, app, and infrastructure behind personalized, safety-critical support for mothers. My background is in modeling people's goals, beliefs, and emotional states, which is the core of how Elevare personalizes care.

Experience

Founding Engineer · Elevare Health

Apr 2026 – Present
Clinical AI infrastructure for maternal care continuity.
  • Lead engineering across a team of 4 full-time onshore engineers and an offshore development team; own the technical roadmap, milestones, and the engineering org's SOPs and tooling, including GitHub org setup, onboarding, and project management.
  • Architected Vera, Elevare's AI agent for mothers, as a LangGraph multi-agent system (conversational agent, routing agent, summarizer), including its safety-critical escalation architecture that routes clinical concerns to human providers (human-on-the-loop).
  • Designed Vera's longitudinal intelligence layer, combining personalized anomaly detection over wearable data with a long-term patient memory system to surface early risk signals.
  • Built the backend on async SQLAlchemy and a 20-table PostgreSQL schema, with HIPAA technical controls (KMS envelope encryption, audit logging, PHI redaction); own AWS infrastructure and access management and support SOC 2 readiness.
  • Took the patient app from 0 to 1 through TestFlight; now leading a React Native rebuild shipping to iOS and Android.

Applied AI Engineer · Advisor360°

Sept 2025 – Apr 2026
  • Shipped an LLM + RAG Slack triage system to reduce ticket time-to-resolution and overall volume across production and non-production workflows.
  • Built an auto-assignment agent with ~80% routing accuracy, powered by a routing policy and service-to-team mapping derived from developer interviews.
  • Developed an analytics and retrieval agent that summarizes recurring issues and recommends resolutions via embedding-based matching to past tickets.

Machine Learning Engineer · MITRE

May 2022 – Jun 2023
  • Automated the NIIRS imagery evaluation workflow in Python, replacing manual scoring with a reproducible ML pipeline.
  • Built a predictive model of tracker performance from motion-imagery quality factors; published at IEEE AIPR.
Research

Doctoral Research · Northeastern University

Sept 2023 – Present
CESAR Lab, advised by Stacy Marsella. Focus: AI systems that learn accurate models of the people they work with.
  • Built a capability-elicitation framework that helps AI teammates learn what a human partner can actually do, improving human–robot team performance (first author, HAI 2026).
  • Built the GAMBiT Theory-of-Mind defender agent, which infers a cyber attacker's cognitive biases from behavior; validated with 61 participants in a simulated enterprise network.
  • Co-designed a multi-agent LLM framework for social-skills training through improvised role-play (IVA 2026).
Education

Industry Ph.D. Candidate in Computer Science

Expected May 2028
Northeastern University, Khoury College of Computer Sciences · Khoury Distinguished Fellow

M.S. in Computer Science

Sept 2023 – May 2025
Northeastern University, Khoury College of Computer Sciences

B.S. in Data Science & Cognitive Psychology

May 2023
Simmons University · Summa Cum Laude · Honors Program · Minors in Computer Science and Spanish
Selected Publications
  1. Hirschmann, S., & Marsella, S. (2026). Aligning Teams across the Human–Artificial Divide: Optimizing Collaboration through Strategic Implicit Communication. HAI 2026.  First author
  2. Beltz, B., et al. (incl. Hirschmann, S.) (2026). Guarding Against Malicious Biased Threats (GAMBiT): Experimental Design of Cognitive Sensors and Triggers with Behavioral Impact Analysis. Computational Brain & Behavior.
  3. Dincer, B., Hirschmann, S., & Marsella, S. (2026). Towards a Virtual Human Improv Troupe for Social Skills Training. IVA 2026.
  4. Hirschmann, S., & Marsella, S. From Implicit to Encoded: How LLM-Based Ticket Routing Reshapes Coordination Practices in a Distributed Engineering Team. Under review, CSCW 2027.  First author
  5. Hirschmann, S., et al. (2022). The Relationship Between Tracking Performance and Video Quality. IEEE AIPR.  First author
  6. Plus 4 arXiv preprints (2025) on LLM-based inference of cyber attackers' cognitive traits. Full list on Google Scholar.
Skills
Languages & tools: Python · SQL · JavaScript · Swift/SwiftUI · React Native · R · Git · Firebase
Backend & cloud: PostgreSQL · async SQLAlchemy · AWS (Fargate, RDS/Aurora, Bedrock, Cognito, KMS, IAM Identity Center) · event-driven architecture (transactional outbox)
AI / LLM: multi-agent orchestration (LangGraph) · RAG & embedding-based retrieval · LLM evaluation & guardrails · agent memory design (temporal event graphs) · Theory of Mind & POMDP user modeling · Bayesian inference · time-series anomaly detection
Health tech & compliance: HIPAA technical controls (envelope encryption, audit logging, PHI redaction) · SOC 2 · wearable data pipelines · vendor BAA evaluation
Product & leadership: iOS/Android delivery (TestFlight, App Store Connect, Play Console) · engineering team leadership · technical roadmapping · experimental design & A/B testing
Spoken: Bilingual English / Spanish