IEEE-published STEM student researcher, inventor, and developer building AI systems for education, security, and disaster resilience.
Research at George Mason University and the Indian Institute of Technology Roorkee. Two USPTO provisional patent applications. Founder and director of Turbo Tutors, a 501(c)(3) nonprofit.
Academic Foundation
Rigorous STEM coursework, strong test scores, and university-level study completed during high school.
South Brunswick High School, NJ
Class of 2027, Weighted GPA 4.4
4.4Weighted GPA
16AP / Honors Courses
2University Courses
1560
SAT Total
Math 800, Reading & Writing 760
Coursework
STEM · AP · Honors, 2023–2027
Every STEM, Advanced Placement, and Honors course on my transcript, by year. Highlighted tags are AP or college-level.
Grade 92023–24
AP Physics 1Honors Algebra IIComputer Science in the 21st Century
Grade 102024–25
AP Calculus ABAP Computer Science AHonors Chemistry IHonors Biology IPre-Calculus A/B · UT AustinHonors English IIHonors U.S. HistoryHonors Spanish III
Grade 112025–26
Multivariable Calculus & Linear AlgebraAP Physics CAP English Lang & CompCS Academy: VR & Game DesignCS Academy: Mobile App DevelopmentHonors Global Studies
Grade 122026–27In Progress
Differential EquationsAP StatisticsAP CybersecurityAP English Lit & CompCS Academy: Data StructuresCS Academy: Computer Science CapstoneHonors Public Speaking
Additional college-level & online
CS50, HarvardX / edXCornell Tech: Interacting with AICornell Tech: Future Builders Studio
Research Mentors
George Mason U.Prof. Mihai Boicu
IIT RoorkeeProf. Ritesh Kumar
Pre-College Program
Jul 2026
Cornell Tech, Cornell University
Summer Innovation Intensives, New York, NY, 2 Cornell University credits
Completed TECHIE 1111: Interacting with AI and TECH 1900: Future Builders Studio. Studied how AI shapes human behavior, communication, and agency, and applied user-centered, responsible-AI design to build and present Vouchify, a reference-verification platform, at the program showcase. Also created a research poster on intelligent tutoring systems, designing and running surveys to collect the original data behind its findings.
Clubs & Athletics
Clubs
FBLAState Representative
MathletesMember
DECA, HOSA, XSTEM
Sports
TaekwondoBlack Belt
Squash
Track & Field
Publications & Research
Three peer-reviewed publications, one research internship, and ongoing independent research. Indexed in IEEE Xplore and presented at IEEE MIT URTC 2025.
Featured Publication
IEEE XplorePublished in IEEE Xplore, May 26, 2026
Developing a Socratic Evaluation Software using Argumentation and Graph-based Open Student Modeling
Indexed in IEEE Xplore. Presented at IEEE MIT URTC 2025 (Undergraduate Research Technology Conference), MIT, Cambridge, MA. Mentor: Prof. Mihai Boicu, George Mason University
Argumentation and Graph-based Open Student Modeling Software (ARGOS) is an Intelligent Tutoring System that builds a dynamic Open Student Model from real-time conversation. Earlier methods model student understanding, but none hold a live Socratic dialogue while continuously updating a quantitative picture of what the student knows. This paper presents two pre-experiments that shaped the system.
01Choosing the tutor
Tested ChatGPT 4o, Claude Sonnet 4, Grok 3, Gemini 2.5 Pro, and Meta AI as Socratic tutors. Gemini 2.5 Pro scored highest (26.625/30) on a rubric for educational quality, factual precision, and instructional adherence.
02Grading understanding
Gemini and four researchers graded eight mock conversations in math and history. Gemini tracked the human graders closely (r = 0.837) but graded more strictly and drifted further from human patterns in the humanities.
03The system
A knowledge-tracking engine powered by the Gemini 2.5 Pro API scores learning progress through rubric-based assessment of Socratic dialogue, built on a React front end, a FastAPI back end, and a Firestore database.
A Comparative Evaluation of Large Language Models for Integration within Argumentation and Graph-based Open Student Modeling Software (ARGOS)
Journal of Student-Scientists' Research, published Sep 25, 2025. Section: College of Engineering and Computing: Department of Information Sciences and Technology. Mentor: Prof. Mihai Boicu
Evaluated ChatGPT 4o, Gemini 2.5 Pro, Grok 3, Meta AI, and Claude Sonnet 4 as Socratic tutors for ARGOS. Each model guided an imaginary student factoring 12x² + 17x + 6 through four standardized scenarios: a correct solution, a conceptual error, a calculation mistake, and going off-topic. Four researchers scored every conversation on a 30-point rubric for educational quality, factual accuracy, and instruction following. Gemini 2.5 Pro scored highest (mean 26.625, SD 2.55), and a one-way ANOVA showed the differences were statistically significant (p = 3.05 × 10⁻⁷), so ARGOS is built on Gemini.
A Comparative Study of Gemini 2.5 Pro and Human Evaluative Metrics in Assessing Student Comprehension Across STEM and Humanities Topics through Conversation-based Assessments
Journal of Student-Scientists' Research, published Sep 25, 2025. Section: College of Engineering and Computing: Department of Information Sciences and Technology
Tested whether Gemini 2.5 Pro can grade student mastery the way people do. Eight simulated conversations, four in algebra and four in history, were given to Gemini line by line with a 100-point rubric covering accuracy, clear reasoning, conciseness and clarity, relevance, and effort, and four researchers scored the same conversations independently. Gemini's scores correlated strongly with the human averages (r = 0.837) but ran about 10% lower and varied more (SD = 14.66). It handled step-by-step math well but often missed partial understanding in open-ended history answers, pointing future versions of ARGOS toward graph-based models of student thinking.
Research Intern, ECG Lab, Department of Earthquake Engineering
Machine Learning for Seismic Risk Prediction and Insurance Premium Optimization
ECG Lab, Department of Earthquake Engineering, Indian Institute of Technology Roorkee. Mentor: Prof. Ritesh Kumar. Online research internship, Jul 1 to Sep 7, 2025.
Trained machine learning models on past seismic activity to predict future earthquake risk, for use in architectural insurance premium calculation. Collaborated with a PhD student at the ECG Lab on the development and analysis of the predictive models, handling data preprocessing, model development, and interpretation of results.
Trained machine learning models on recorded seismic activity to estimate future earthquake risk
Performed data preprocessing, exploratory analysis, and interpretation of earthquake-related datasets
Applied the resulting risk models to architectural insurance premium calculation
Collaborated with a PhD student at the ECG Lab, Department of Earthquake Engineering
Digital Surveillance & Human Rights
Mar 2026 – Present
Independent Research
Researching the impact of digital surveillance technologies on privacy, freedom of expression, and civil liberties. Examining how monitoring systems influence human behavior and evaluating policy frameworks that balance public safety with fundamental human rights.
Investigating the "chilling effect," where individuals alter speech, online activity, and information-seeking behavior when they believe they are being observed
Analyzing real-world case studies involving digital surveillance, electronic device searches, social media monitoring, and public-health tracking systems
Evaluating legal, ethical, and policy frameworks designed to balance technological capabilities with transparency, accountability, privacy, and civil liberties
Patents
Two provisional patent applications filed in 2026.
Unified Disaster Intelligence and Adaptive Response System
Inventor: Ansh Saini, 35 U.S.C. § 111(b)
A live flood, wildfire, and earthquake map with explainable risk scoring and voice-guided evacuation routing. Every risk number is built from named, inspectable signals and shows the reasoning behind it. Extends FloodPilot into a unified multi-hazard platform. Shared with the New Jersey State Climatologist for review of its weather data and thresholds.
Global Strategic Straits Intelligence and Resilience System
Inventor: Ansh Saini, 35 U.S.C. § 111(b)
Simulates Hormuz-style chokepoint closures and the trade fallout that follows. Covers the Strait of Hormuz, Malacca, the Suez Canal, Bab el-Mandeb, and the Panama Canal, estimating effects on shipping routes, trade, and environmental risk through a digital-twin dashboard.
Founding and running a free tutoring nonprofit, meditation-education outreach, and wildfire relief fundraising.
Turbo Tutors, 501(c)(3) Nonprofit
Early 2024 to present
500+Hours of free tutoring
100+Students served
10+Countries reached
15+Tutors trained
Founder and Director. I started Turbo Tutors as a freshman and have run it since, delivering 500+ hours of completely free tutoring to more than 100 students in 10+ countries. It is a federally recognized 501(c)(3) charitable organization.
Trained a team of 15+ volunteer tutors
Led a team of five to build turbotutors.org, the student registration system, and tutor onboarding
Meditation has changed how I live and work, and since 2023 I have built my life around my Heartfulness practice. The patience it has taught me has carried me through months of AI research, two patent applications, and building TurboTutors from the ground up.
About MeHeartfulness practitioner since 2023
Heartfulness Institute
Sep 2023 to present
200+Volunteer hours
250+Schools reached
2×Gold PVSA awards
Outreach and event planning volunteer since Sep 2023, contributing more than 200 hours. I led meditation sessions and ran meditation-education outreach that reached over 250 schools nationwide, including my own high school, building the school contact databases behind it and coordinating weekly gatherings that served hundreds of participants.
Led meditation sessions and coordinated weekly community gatherings
Built the school contact databases that drove outreach to 250+ schools across the United States
Ran Heartfulness events at my own high school, including informational booths
Recognized with two Gold President’s Volunteer Service Awards, the highest tier awarded
Organized and ran an official American Red Cross donor drive for California wildfire victims, raising $5,015 from 59 donors and surpassing the $5,000 campaign goal. Rather than only soliciting donations, I taught a paid Python coding summer camp for younger students and donated every dollar of the proceeds, building the fundraiser around a skill I could actually offer.
Designed two fundraising flyers and a wildfire campaign poster
Set up in-person donation tables at a local gym and the Princeton Club
Ran a Coding for Kids Python summer camp for ages 10 to 14, with all proceeds donated
01Coding for Kids summer camp flyer, all proceeds donated
02Wildfire relief campaign poster
03Fundraiser flyer
04Donation table at a local gym
05Donation table at the Princeton Club
Internships & Experience
Research internships at George Mason University and IIT Roorkee, plus industry experience in AI/ML and STEM education.
Research Internships
2025
Jul to Sep 2025, Fairfax, VA
ASSIP Research Intern, George Mason University
Department of Information Sciences and Technology, College of Engineering and Computing
Conducted two studies supporting ARGOS: evaluated five LLMs as Socratic tutors and compared AI-generated assessments of student understanding with human evaluations. The research resulted in two ASSIP publications and an IEEE Xplore–published conference paper.
Jul 1 – Sep 7, 2025, Remote
Research Intern, ECG Lab, Department of Earthquake Engineering, IIT Roorkee
Project: Machine Learning for Seismic Risk Prediction and Insurance Premium Optimization. Mentor: Prof. Ritesh Kumar, PhD. Collaborated with a PhD student at the ECG Lab on data preprocessing, model development, and interpretation of results.
Industry Experience
Summer 2024
Jul–Aug 2024, Remote
AI/ML Engineering Intern, Dashworks AI
Only high schooler on the team. Self-taught the ML libraries (Hugging Face, PyTorch), helped build an LLM and RAG enterprise chatbot, and trained and optimized model weights.
Jul 2024, South Brunswick, NJ
STEM Instructor, Code Ninjas
60 hours. Ran coding and engineering classes of 15+ students, helped restructure curricula, and adapted lessons to student needs.
Aug 5–31, 2024, Princeton, NJ
STEM Camp Facilitator, STEAM Works Studio
Paid position. Ran "Crazy Chemistry" and "Little Doctors" camps with 20+ students per session and coached students through their first engineering projects.
Projects
AI/ML, disaster response, maritime intelligence, financial literacy, and educational technology.
Featured Projects
10
UDIARS
Winner
Best User Experience Winner, Youth Code x AI Hackathon, May 2026
The Unified Disaster Intelligence and Adaptive Response System (UDIARS) combines flood, wildfire, and earthquake data in one application so users can review local hazard conditions and compare possible evacuation routes.
Connected public data feeds from NOAA, NASA FIRMS, USGS NWIS, and USGS Earthquake services
Built separate risk scoring methods for flood, wildfire, and earthquake hazards
Used FastAPI to combine the hazard data and React to display maps, risk levels, and route options
Tested the application with different locations and hazard scenarios
Presented the project at the Youth Code x AI Hackathon, where it received the Best User Experience award
CyberShield analyzes security telemetry across digital and cyber-physical systems, classifies threats with Gemini structured output, scores real-world risk, checks CVEs against CISA KEV, and generates explainable remediation reports for human review.
Classified threats with Gemini structured output and mapped findings to MITRE ATT&CK, MITRE ATLAS, and OWASP LLM Top 10
Built a deterministic cyber-physical risk score using severity, criticality, autonomy, safety relevance, exposure, and human oversight
Separated digital and physical impact so risk changes based on the asset under review
Checked CVEs against CISA's Known Exploited Vulnerabilities catalog
2nd Place Winner, AceSAT Education AI-Agent Hackathon, Aug 2026
Adaptly is an adaptive SAT preparation agent that learns from a student's scores, practice, and mistakes, decides what to work on next, tutors them through targeted practice, and adapts automatically as they improve.
Built for students like those I tutor at Turbo Tutors
Built a model that tracks scores, practice history, and mistakes to estimate strengths and weak areas
Added an adaptive engine that selects what to work on next based on recent performance
Developed targeted practice with step-by-step tutoring and explanations for missed questions
Designed the system to update its recommendations automatically as the student improves
Organized progress tracking so students can see how their preparation changes over time
1st Place Winner, LUMA Pitch Competition, Teen Financial Literacy, Jul 2026
StartSmart is a free interactive web application that helps high-school students build practical money-management skills through short lessons, calculators, quizzes, and real-world examples.
Created 12 learning modules covering budgeting, banking, credit, taxes, renting, debt, investing, and financial safety
Built calculators for budgeting, taxes, APY, rent affordability, and debt to income ratio
Developed a financial glossary
Added a 20 question quiz for reviewing key concepts
Organized the application for independent, self paced learning
Top 100 of 2,120 participants, Reverie Hacks 2026, Adaptive AI Textbook & Collaborative Learning Platform, Aug 2026
DigiBook turns PDF, DOCX, and TXT materials into an adaptive learning experience with AI tutoring, personalized practice, accessibility tools, note-taking, and live group study.
Built for students like those I tutor at Turbo Tutors
Selected in the Top 100 among 2,120 participants at Reverie Hacks 2026
Built multilingual Socratic tutoring, quizzes, flashcards, concept maps, and an open student model
Added an Adaptive Learning Path that uses quiz results and learning activity to recommend what to study next
Built a Study Notebook with typed or dictated notes, source-linked excerpts, and a drawing/sketch canvas
Added live Study Rooms with shared reading, collaborative notes and sketches, chat, and audio/video communication
Designed accessibility for blind and low-vision, deaf and hard-of-hearing, dyslexic, motor, and attention-diverse learners through Read Aloud, screen-reader and keyboard support, voice controls, high contrast, adjustable text/spacing, reading ruler, and reduced motion
Applied ideas from the ARGOS intelligent tutoring research
George Mason University ASSIP, Mentor: Prof. Mihai Boicu, Jul to Sep 2025
Argumentation and Graph-based Open Student Modeling Software (ARGOS) is an intelligent tutoring system that uses Socratic dialogue, argumentation theory, and knowledge graphs to represent student understanding. My research focused on evaluating its language model tutoring and assessment components.
Evaluated five large language models as Socratic tutors using a structured 30 point rubric
Compared model performance across reasoning quality, instructional effectiveness, and consistency
Compared AI generated assessments of student understanding with human evaluations
Analyzed differences across the models and agreement between AI and human ratings
Contributed to research that resulted in two ASSIP publications and one IEEE conference paper
My role: Evaluated five language models, applied the 30-point scoring rubric, analyzed model performance and AI and human rating agreement, and contributed to the publications and conference presentation.
The Global Strategic Straits Intelligence and Resilience System (GS-SIRS) helps users examine risk at major maritime chokepoints and explore how closures, delays, or rerouting could affect shipping and connected systems.
Added strategic chokepoints including Hormuz, Malacca, Suez, Bab el Mandeb, and Panama
Organized vessel activity and disruption indicators into a risk scoring workflow
Created scenarios involving closures, delays, and vessel rerouting
Built a digital twin dashboard to display possible effects on transportation, trade, economic activity, and the environment
Tested different disruption scenarios across multiple chokepoints
USAII Global AI Hackathon 2026, Round 2, Team Project, Jun 2026
Triage is a web application that interprets a user's description of an immediate need, estimates its urgency, and identifies relevant nearby support programs.
Used Gemini 2.5 Flash to classify user needs and estimate urgency
Connected Google Places to locate nearby services
Added checks to remove programs that appeared closed or unavailable
Organized the results into categories and possible next steps
Tested the application with different housing, food, health, and safety situations
My role: Helped design the needs classification workflow, develop and test the application, and prepare the final demonstration.
Vouchify is a structured reference platform that helps students and early career candidates demonstrate strengths that traditional résumés often miss. Candidates request feedback from teachers, mentors, coaches, and supervisors, which is organized into a shareable professional profile for employers.
Designed separate workflows for candidates, reference providers, and employers
Organized feedback around reliability, communication, teamwork, initiative, and work ethic
Added consent, privacy, and reference management controls
Conducted early user and employer outreach to understand interest and concerns
Co developed and presented the prototype at the Cornell Tech Future Builders Studio showcase
My role: Product research, workflow design, market validation, prototype development, user outreach, and final presentation.
FloodScore combines FEMA flood zone information and USGS seismic data to help homeowners and real estate buyers explore possible property risk and long term financial exposure.
Connected FEMA National Flood Hazard Layer information with USGS seismic data
Built a property risk scoring method
Created calculations for possible financial exposure over a 30 year period
Displayed the results through a property focused map and dashboard
Prepared the project for submission to HackAmerica
GatewayGS, AI 4 Earth Hackathon, Planetary Cascade Intelligence, Aug 2026
Built an AI-assisted planetary cascade intelligence platform that models how regional environmental stress can propagate through connected infrastructure, resource, and human systems, helping decision-makers understand both local pressure and wider systemic impact.
Integrated current sea-temperature, atmospheric, and air-quality signals with transparent regional risk modeling
Developed seven environmental and human-pressure domains to calculate Regional Stress Score and Global Cascade Potential
Built a dependency-graph model that traces cascading impacts across infrastructure, resources, human systems, and documented cross-region connections
Created a scenario simulator that reruns the full model under conditions such as ocean warming, data-center demand growth, and infrastructure disruption
Implemented a constrained Gemini AI strategist that selects and explains pre-verified intervention portfolios while deterministic models calculate and verify all impact scores
Hacksocial 2026, Recovery Planning and Resource Support, Aug 2026
LifeBridge is a non-emergency planning application that helps users organize their needs following a difficult life event. It categorizes user-entered information across housing, food, health, finances, and safety and connects users with publicly available resources.
Built an intake workflow covering housing, food, health, finances, and safety
Created a LifeBridge Score to organize the needs entered by the user
Developed a recovery checklist with progress tracking
Connected users with location based public resources
Kept emergency and crisis contact information visible when urgent help may be needed
RevenueCat Shipaton 2026, College Search and Application Planner, Oct 2026
Matricula is a college-planning web app that helps students explore schools and plan applications using official, authoritative data instead of scattered or unverified sources.
Organized official college data into a searchable planning interface
Built tools to compare schools across admissions, cost, and outcome information
Helped students track and plan their application steps in one place
Focused on transparent, source-based information for college decisions
Released a free web version alongside an iOS build with RevenueCat-powered subscriptions for the Shipaton hackathon
The Cornell Tech program, press coverage of my research, and the practices I keep up outside of school.
Cornell Tech Summer Innovation Intensives
Jul 13–30, 2026 · New York, NY
Selected for Cornell Tech's Summer Innovation Intensives, earning 2 Cornell University credits across Interacting with AI and the Future Builders Studio. Built and presented Vouchify, a reference-verification platform, at the program showcase.
01Presenting Vouchify at the Cornell Tech Future Builders Studio showcase
My IEEE-published research was covered regionally by MyCentralJersey, syndicated nationally by AOL News, and reported locally by TAPinto South Brunswick.
Regional · MyCentralJersey · Jun 17, 2026
NJ junior publishes AI study that could change how students learn
Regional coverage by MyCentralJersey of my IEEE-published AI research on student learning and my presentation at IEEE MIT URTC. By Jack Salaki.
SBHS Student Ansh Saini has a Research Paper from a GMU Internship Presented at MIT
Coverage of the research paper from my George Mason summer internship, presented at IEEE MIT URTC 2025 and published in IEEE Xplore. “The greatest impact of technology is realized when it is deployed in service of others.”
SBHS Student Rallies Community to Support California Wildfire Victims
Recognition of the American Red Cross fundraiser I led for California wildfire victims. “Your hard work, kindness, and leadership made a big difference.”
Beyond academics, research, and building, I make time for practices that strengthen focus, consistency, and balance. I continue my Heartfulness journey through meditation, community service, and the PEARL (Practical, Educational, Actionable, Rejuvenating Lifestyle) webinar series. Running has become another part of that journey, including training for and racing 5Ks.
HeartfulnessMeditation · PEARL learning · Community service
Running
CityTri, New Jersey Summer Half, 5K
Monmouth Battlefield State Park · Englishtown, NJ
Princeton Battlefield 5K
Princeton Battlefield State Park · Princeton, NJ
CityTri result & race video
🥉 3rd Place, 15–19 Age Group · New Jersey Summer Battlefield Run