← BACK TO CAREERS
Research Scientist Intern
San Francisco Bay Area
·
Full-time
·
$65 - $80/hour
·
Master’s or PhD
Nirva is building a long-term, context-aware conversational system that helps people better understand patterns in their communication, emotions, relationships, and behavior. We are looking for a full-time Research Intern to help define and measure what emotionally intelligent, trustworthy AI behavior should look like. This is an opportunity to build a publishable research contribution at the intersection of affective computing, human-AI interaction, agent evaluation, and product design. You will work in person in San Francisco with Nirva’s Chief of Staff, Founders, and ML research leadership. The initial internship term is four months, with a flexible start date and potential extension based on research progress, mutual fit, and company needs.
ABOUT NIRVA
Nirva is building the first AI fashion accessory for your mind — a necklace or bracelet that privately listens only to the wearer to enable proactive journaling, emotion tracking, and life coaching. Think Oura Ring for your Mind.
Backed by Lightspeed and South Park Commons, recognized at CES within 90 days of founding, and launching in the U.S. this fall. Co-founders bring backgrounds from Meta Reality Labs (Quest, Ray-Ban glasses) and fashion industry.
THE ROLE
Many AI systems can sound empathic. Far fewer can accurately understand a user’s emotional context, acknowledge uncertainty, offer useful pushback, preserve user agency, and remain trustworthy over time. Your work will help establish a rigorous answer to the question: What does emotionally intelligent, calibrated, and safe behavior look like for a conversational AI agent? You will help create both: a public benchmark for evaluating emotional intelligence and trustworthiness in chatbot and agent interactions; and an internal evaluation suite that helps Nirva improve its conversational system, memory behavior, recommendations, and product guardrails. We are especially interested in evaluation that goes beyond generic “empathy” or emotion-recognition tests. The goal is to measure whether an AI system can be emotionally perceptive, context-sensitive, honest without being harsh, appropriately uncertain, and respectful of the user’s autonomy.
WHAT YOU'LL OWN
What you’ll do
Review and synthesize research across emotional intelligence, affective computing, social cognition, moral psychology, human-AI interaction, AI safety, and LLM/agent evaluation.
Develop a clear, testable taxonomy for emotionally intelligent AI behavior.
Design benchmark tasks, scenario templates, rubrics, annotation guidelines, and evaluation protocols.
Build and analyze model baselines, including out-of-the-box models and prompt-engineered or agentic systems.
Help develop evaluation methods for multi-turn and longitudinal chatbot-user interactions.
Run empirical studies and analyze failure modes in emotional reasoning, calibration, constructive honesty, user agency, and safety.
Translate research findings into concrete recommendations for Nirva’s chat experience, prompting, memory use, and ethical product principles.
Write a research manuscript and prepare the work for submission to an appropriate venue and/or public preprint server.
WHAT WE'RE LOOKING FOR
Required
You are currently pursuing, or have recently completed, a Master’s or PhD degree.
You can work full-time and in person in San Francisco.
You are a strong researcher and writer who can turn ambiguous human concepts into measurable constructs, experiments, rubrics, and concrete evaluation criteria.
You are comfortable with Python and quantitative analysis.
You can independently drive a research project while collaborating closely with engineers, founders, and product leaders.
Preferred
Your thesis, dissertation, coursework, or prior work relates to emotional intelligence, affective computing, cognitive science, psychology, philosophy, HCI, social computing, conversational AI, moral reasoning, AI alignment, or AI evaluation.
You have authored or materially contributed to a paper, poster, thesis, research report, or other substantial research writing.
You have experience with LLM or agent evaluation, human preference studies, annotation design, survey design, psychometrics, experimental design, or qualitative research.
You are familiar with foundational emotional-intelligence research, including work by Peter Salovey, John D. Mayer, David Caruso, Daniel Goleman, or related researchers.
You have worked with prompt-engineered, tool-using, memory-enabled, or multi-turn agent systems.
We are particularly excited about candidates who pair technical rigor with deep curiosity about human behavior. This role is not only about maximizing model capability; it is about understanding how AI should behave when people trust it with emotionally meaningful conversations.
COMPENSATION
Full-time · In person · San Francisco · Flexible start date · $65 - $80/hour
The initial internship term is four months, with potential extension based on research progress, mutual fit, and company needs.
Mentorship and resources
You will work directly with Nirva’s Chief of Staff and founding team, with technical mentorship from Su Zhang, a Machine Learning Engineer at Nirva Labs with a Ph.D. in Computer Science and a research background in reinforcement learning and applied machine learning.
Nirva will provide the resources needed to execute serious work, including technical collaboration, compute and experimentation support, annotation and research budget as appropriate, and regular feedback from product and engineering leadership.
We do not treat research as an internal slide deck. The goal is a meaningful and publishable contribution.
TO APPLY
A HUMAN WILL READ EVERY SINGLE APPLICATION
