Sarang Goel

I'm a former computer science + mathematics student at Stanford and co-founder of Triage.

At Triage, we're working to close the gap between what AI is capable of and what it can reliably do in practice. We're building infrastructure that lets it take on more work without needing constant human attention.

Experience

Triage

Co-Founder

San Francisco, CA

  • Building infrastructure that helps AI systems generalize to unfamiliar environments by incorporating guidance from experienced humans.
  • Raised $1.5M in pre-seed funding at a $12M valuation, backed by BoxGroup, Precursor, Thomas Dohmke, Zach Lloyd, and others.

Point72

AI Engineer Intern

New York City, NY

  • Implemented a context management system for internal analysis agents, combining offline schema profiling and confidence-scored key and join inference with graph-based retrieval and compression of dataset metadata, lineage, metric definitions, and validated SQL examples.
  • Developed SQL/Python simulation and analysis workflows that maintained multi-turn state, enforced user permissions and grain-aware joins, bounded execution costs and query-repair attempts, and reconciled outputs against trusted metrics to catch semantic errors.
  • Built a dashboard platform that presented execution plans, selected datasets, filters, time windows, and assumptions alongside generated queries and supporting artifacts, enabling trading pods to inspect agent decisions and trace results back to source data.

Amazon

Software Engineer Intern

Santa Cruz, CA

  • Built a multi-agent pipeline for real-time driver hazard detection from images, audio, and video using vision, audio, context, and risk agents.
  • Developed a projection framework to distill teacher-scale vision–language embeddings into a lower-dimensional student space for retrieving and mapping delivery locations from visual and textual inputs.
  • Built event-driven and batch ingestion infrastructure for 10M+ daily delivery stops, with geospatial hazard flagging and automated driver alerts.

MIT Media Lab

Embedded Systems Research Intern

Cambridge, MA

  • Conducted research through the Research Science Institute (RSI), which selects 100 students worldwide.
  • Built an AI-powered wearable for real-time detection of Freezing of Gait in Parkinson’s patients, using soft vibrations to support normal gait and directed vibrations to help overcome detected episodes. (Presentation)

Quantitative Imaging and AI Lab, Stanford University

AI Research Intern

Stanford, CA

  • Developed and published an automated region-of-interest selection method for segmenting hyper-reflective foci in OCT images using a ResNet34 patch-based model with binary cross-entropy loss.
  • Trained a domain-specific generative language model on ophthalmology literature to aid medical diagnosis.

Smart Electronic Systems Lab, UNT

Computer Vision Research Intern

Denton, TX

  • Developed and published NuTRILog for estimating food nutritional value, combining SSD EfficientNetB0 with FPN for food classification and localization and OpenCV card-edge detection for size estimation.

Projects

ProtoGen: A Hardware Prototyping and Design Automation Platform

Built an RL-based hardware prototyping platform that streamlines hardware development from concept to production. The pipeline extracts design requirements via interactive questioning, selects components under optimization constraints, and trains policies with simulator-in-the-loop rewards to auto-generate and refine wiring schematics, exporting production-ready design files.

ProtoGen: A Hardware Prototyping and Design Automation Platform project image

Method for Determining Recirculation Mode in Vehicles

Invented and self-patented a sensor-less method to regulate air quality in vehicles, utilizing camera-based traffic analysis to close the air inlet valve during toxic pollution conditions and reopen it based on high CO2 buildup in the vehicle cabin space.

Method for Determining Recirculation Mode in Vehicles project image

Therms: Thermoregulation Wearable with AI-powered Insights

Therms combines advanced thermoelectric technology with a deep understanding of human physiology to create a wearable system that actively supports thermal homeostasis. By targeting the glabrous skin on the feet, Therms achieves efficient, non-invasive thermoregulation while integrating AI-powered insights to improve health outcomes and performance.

Therms: Thermoregulation Wearable with AI-powered Insights project image

Course projects

Sequential Reinforcement Learning on Economic Discourse for Real-Time Financial Trading

CS224R · Paper · Poster

SkyBites - Reinforcement Learning for Real-Time Drone Fleet Coordination in Campus Food Delivery

CS238 · Paper

Integrating Relational Message Passing and Knowledge Graphs in LLMs for Component Recommendation

CS224N · Paper · Poster

Linear Probing of Physical Plausibility Representations in Diffusion Image Models

CS229 · Paper · Poster

Subset-Consistent Latent Reconstruction for Robust RGB-D Segmentation

CS231N · Paper

Forecasting of COVID Cases in U.S. Cities: Integrating Mobility, Vaccination, and Policy Data

DATASCI112 · Poster

Awards