Applied Scientist · Ph.D.

Sara Malvar

Sara Malvar SM
  • 10+years in AI & data
  • 300+mentees
  • 14+patents

About

Applied Scientist specializing in AI systems, large language models, and evaluation frameworks. My work focuses on building the infrastructure, methodologies, and quality systems that enable enterprise AI agents to be measured, optimized, and deployed with confidence. I lead the design of evaluation, benchmarking, and grading platforms used to assess model quality, reliability, and alignment across real-world applications.

With 10+ years of experience and a Ph.D. in Applied Sciences, I've worked across academia and industry, translating research into production-scale AI systems spanning enterprise AI, natural language processing, geospatial intelligence, environmental modeling, and machine learning platforms. I'm an inventor on 14+ patent applications and author of multiple peer-reviewed publications in AI, machine learning, and large language models.

Beyond technical contributions, I've mentored 300+ students and professionals in data science and machine learning, and regularly contribute to discussions on AI evaluation, benchmarking, and trustworthy AI. My interests lie in building reliable AI systems that bridge scientific rigor, product impact, and real-world adoption.

Latest projects

Microsoft 365 · Frontier Tuning

Teaching enterprise AI agents to work the way teams do

Frontier Tuning is Microsoft's approach to adapting AI agents to a company's data, processes, and workflows inside its compliance boundary. My work focuses on the evaluation, benchmarking, and grading systems that measure whether tuned agents are getting better across RFT, SFT, and test-time optimization workflows.

  • Evaluation backbone for Copilot agent workflows.
  • Grading systems for reinforcement and supervised fine-tuning.
  • LLM-as-a-judge, calibration, hallucination detection, and trajectory scoring.

Microsoft Research · FarmVibes.AI

Multi-modal geospatial AI for agriculture and sustainability

FarmVibes.AI helps researchers and practitioners fuse satellite imagery, drone imagery, weather, sensor data, and other spatiotemporal sources to build richer insights for agriculture, sustainability, emissions, and soil health.

  • Fusion workflows for geospatial and remote-sensing ML.
  • Data ingestion, preprocessing, and model training workflows.
  • Open-source tools for robust agriculture and sustainability insights.

Publications

  1. 2026
    Diagnosing Capability Gaps in Fine-Tuning Data

    S. Asgari Taghanaki, R. Agarwal, S. Malvar, L. O. Nunes, R. Chandra, E. Kiciman, et al.

    arXiv:2604.27547

  2. 2024
    RAG vs Fine-tuning: Pipelines, Tradeoffs, and a Case Study on Agriculture

    M. A. de Luis Balaguer, V. Benara, S. Malvar, L. O. Nunes, B. Silva, R. Chandra, et al.

    arXiv:2401.08406

  3. 2024
    Injecting New Knowledge into Large Language Models via Supervised Fine-Tuning

    N. Mecklenburg, Y. Lin, X. Li, S. Malvar, B. Silva, R. Chandra, et al.

    arXiv:2404.00213

Full list on Google Scholar.

Patents

Granted

Pending

  • Atmospheric Chemical Species Detection Using Multispectral Imaging

    411320-US · Pending

  • Data-Driven Approaches to Improve Understanding of Process-Based Models and Decision Making

    412147-US · Pending

  • Framework for Analyzing Properties of Chemical Materials

    413559-US · Pending

  • Framework for Language Model Copilot Development

    061428-US01 · Pending

  • Goal-Driven Rubric Generation and LLM Evaluation Framework

    506967-US02 · Pending

  • Interactive Prompting for Supply Chains

    413991-US02 · Pending

  • Large Language Model-Based Document Generation Pipeline

    504866-US02 · Pending

  • Machine Learning Solutions to Predict Protein Characteristics

    412148-US · Pending

  • Pollutant Sensor Placement

    410916-US · Pending

  • Systems and Methods for Emission Source Attribution

    411704-US · Pending

Experience

Download CV (PDF) →

Professional timeline

Microsoft logo
Sep 2024 — Present

Senior Applied Scientist · Tech Lead

Frontier Tuning Evaluation — Microsoft

Technical lead for AI evaluation, benchmarking, and grading infrastructure behind Microsoft 365 Frontier Tuning, Microsoft's platform for fine-tuning enterprise AI agents.

  • Architected the evaluation backbone across Copilot agent workflows.
  • Built grading systems for RFT and SFT.
  • Led LLM evaluation, calibration, hallucination detection, and trajectory-scoring capabilities.
  • Influence product strategy and mentor scientists across Microsoft's AI evaluation ecosystem.
Microsoft Research logo
Aug 2021 — Sep 2024

Sr. Research Software Development Engineer

Microsoft Research

Developed AI platforms and machine learning solutions spanning sustainability, scientific discovery, and enterprise generative AI.

  • Led AI solutions for Dow Chemical, ITC, Unilever, and Land O'Lakes.
  • Built fine-tuning, RAG, and evaluation systems adopted across customer engagements.
  • Contributed to FarmVibes.AI and large-scale geospatial AI platforms.
  • Inventor on patents and author of peer-reviewed AI systems research.
RCGI logo
Jul 2019 — Aug 2021

Postdoctoral Researcher

Research Centre for Gas Innovation

Applied ML across energy, environment, and materials research: materials discovery, NLP for energy-policy sentiment, seismic imaging, sensor data, and bioprocess optimization. Mentored junior researchers.

Udacity logo
2018 — Aug 2024

Data Science & ML Instructor

Udacity · Awari · Alura

Taught and mentored 300+ students and professionals across ML, deep learning, applied LLMs, and 9 Udacity nanodegrees, plus live webinars and cohort-based instruction.

Data Science Dojo logo
2020 — 2024

Data Science & ML Mentor / Instructor

Data Science Dojo

Mentored and instructed learners in Data Science and Machine Learning courses, helping professionals build practical foundations and apply ML concepts with confidence.

University of Tokyo logo
2018

Visiting Researcher

University of Tokyo · SELA Scholarship

Conducted international doctoral research in a highly collaborative academic environment, extending the technical depth and global scope of her applied-science research agenda.

University of Pennsylvania logo
2015 — 2019

Ph.D. Researcher

University of Pennsylvania · University of São Paulo

Advanced machine learning and computational research across applied-science problems, bridging academic rigor with production-oriented modeling practices and cross-institution collaboration.

IBM logo
Pre-2019

Technical Specialist – Data Infrastructure for AI & Analytics

IBM

Worked on applied data science and AI initiatives before Microsoft, building experience across analytics, machine learning, and industry-facing technical problem solving.

Education

University of Pennsylvania logo University of São Paulo logo
2015 — 2019

Ph.D. in Applied Sciences

University of Pennsylvania & University of São Paulo

University of Tokyo logo
2018

Ph.D. Visiting Researcher

University of Tokyo · SELA Scholarship

University of Brasília logo
2014 — 2015

M.Sc. in Applied Sciences

University of Brasília · EMBRAER Best Dissertation Award

University of Brasília logo
2009 — 2013

B.Sc. in Electrical Engineering

University of Brasília · Graduated with honors

Mentoring

I've mentored 300+ students through MentorCruise, Udacity, and Data Science Dojo — helping them go from the fundamentals of Data Science and Machine Learning to confidently applying those skills in projects, research, and industry roles.

  • A personalized learning plan tailored to your goals
  • Mastering Data Science, ML & Generative AI concepts
  • Building portfolio projects that stand out
  • Preparing for research or industry roles
Selected public MentorCruise reviews

Get in touch

Want to collaborate, ask about my work, or start mentoring? I'd love to hear from you.