We build systems that sense people, privately.

The Systems, Networking, and Energy Efficiency (Synergy) Lab in the School of Computer Science at Carnegie Mellon University, led by Prof. Yuvraj Agarwal. We build and deploy real systems at the intersection of hardware, software, and people.

Mites.io ↗ — our full-stack sensing platform: 300+ privacy-preserving multi-sensor devices across TCS Hall, and a living testbed for much of our research.
BuildingChat overview: an occupant asks where to find a comfortable place to work and BuildingChat validates access, selects sensors, retrieves and composes data into an answer.The SpiderGen workflow: generate sample products, coarse process lists, and the final process flow graph, then use it for carbon footprint and supply-chain hot-spot analysis.IMUCoCo estimates full-body pose from IMUs in atypical locations such as a brooch, armband, chest pocket, thigh pocket, or anklet.ECO-LLM architecture: an emulator that explores pipeline configurations offline and a runtime that selects a query resolution path under cost and latency SLOs.Pipeline comparing prior audio featurization with Kirigami's privacy filter, showing fine-tuned speech recognizers recover far less speech after Kirigami while activity recognition accuracy stays similar.Edulyze architecture: drivers ingest data from multiple classroom sensing systems, analytics modules process instructor and student location, gaze, and audio into a unified schema and API.The high-complexity Security & Privacy Facts label for a fictional smart thermostat, with the U.S. Cyber Trust Mark, sensor data practices, and a QR code.Frames from a video: smart thermostat boxes with security and privacy labels on a store shelf, a shopper scanning the QR code, and the detailed label on a phone.Prior work recognizes individual activities like typing or chewing, while TAO combines parallel and sequential activities into contexts such as office work or skipped meals for wellness applications.A multiple price list table asking participants to choose between discount coupons for a smart device with no security updates versus automatic security updates.An example website depending on DynDNS and Cloudflare for DNS and KeyCDN for content delivery.A HelloVisitor app manifest chains push, detect, select, and post operators so an in-home hub extracts faces from camera video and sends only cropped faces to the cloud.
Maps of projected new solar energy generation by U.S. state under a 25% tariff, no incentive, and a 22.9% tax credit.The OrganicHAR framework: identify key moments in sensor data, describe them with a vision language model, and extract activity labels at adaptive granularity.Survey responses on whether participants would use, find reliable, and trust devices with software, hardware, and jamming privacy features.A chart placing computational decarbonization at the meso scale, between device-level decisions and global, decades-long climate policy, across spatial and temporal axes.The Matcha plugin in Android Studio: detected data-access API calls, annotation quick-fixes, generated XML label specification, and a preview of the Google Play data safety label.ThingPoll's negotiation workflow, from profile assignment through privacy queries and configuration suggestions, alongside screenshots of each step in the app.Pipeline of ClassID: within-session ID assignment from a student-facing camera, session-level face and gaze representations, and across-session ID matching using attendance.VAX architecture: pretrained video and audio models label activities to train privacy-preserving sensors, after which the camera and microphone rig is removed for deployment.Mites system architecture: multi-sensor Mites devices in each room stream encrypted data through a gateway to per-building backends serving apps, building managers, and developers.Two storyboards: a sensing system flags which discussion groups stay on topic, and another identifies quiet students for an instructor to call on.TEO workflow: an admin initializes a device, users claim ephemeral co-ownership, the device uploads encrypted data, and requesters must get every co-owner's approval.PSA pipeline: smart apps are parsed into models, combined with device, environment, interaction, and intent models, and checked to produce counterexamples and safe configurations.

Research areas

All research →

Recent projects

All publications →

What's new

All news →
09/2026

Paper on BuildingChat, enabling natural-language interactions with general-purpose sensing infrastructure for buildings, is accepted to Ubicomp'26! Congratulations Prasoon, Anu, Ben, and Haozhe!

05/2026

Paper on The Price to Site, on cost-effective incentives for solar deployment, is accepted to BuildSys'26! Congratulations Cooper and Anu!

12/2025

Paper on OrganicHAR: Activity Discovery in Organic Settings is accepted to Ubicomp'26! Congratulations Prasoon!

12/2025

Professor Yuvraj Agarwal's collaborative work on IoT Privacy and Security Labels with Pardis Emami-Naeini and Lorrie Cranor was awarded the Norm Hardy Prize from the Foresight Institute! [CyLab Release]

10/2025

Paper on SpiderGen: Towards Procedure Generation for Carbon Life Cycle Assessments using Generative AI is accepted to AAAI'26! Congratulations Anu!

Join us

[AY 2026/2027] We are looking for 1 PhD student (starting Fall 2027) and 1-2 stellar M.S. or undergraduate students to join our lab. Our current projects span ubiquitous computing and privacy-preserving sensing, privacy and security for IoT and connected devices, computational decarbonization and sustainable energy systems, and the software and sensing infrastructure for smart buildings. The Research and the Publications tabs on the website summarize the current research directions of our lab. If you are awesome and are motivated to work on the exciting and socially relevant projects at the intersection of hardware and software systems, contact Yuvraj by email. If you are a PhD student, there are multiple programs at CMU that you can apply to and work with us. If you are already at CMU, please drop by Yuvraj's office.

Email Yuvraj →
Our research is supported by

Find us at TCS Hall 235, 4665 Forbes Avenue, Carnegie Mellon University, Pittsburgh, PA 15213 · Map