Hello, I'm

Agastya
Munnangi

I build  |

Computational Biology Researcher · AI Founder · Lambert High School, Class of 2028

Scroll

Bridging Biology,
AI, and Real Impact

I'm a rising junior at Lambert High School in Suwanee, GA — building AI products, doing computational biology research, and competing at the state level in healthcare and business competitions.

From training ML models for cancer therapy prediction at Harvard-tested research platforms to founding Iddu AI — an autonomous AI wellness scheduling app currently in Apple TestFlight — I work at the intersection of biology and artificial intelligence to solve real healthcare problems.

4.0 Unweighted GPA
1530 SAT Score
110+ Tutoring Hours
$10K+ Startup Funding
Lambert High School Suwanee, GA · Class of 2028
University of Miami Computational Genetics Intern
Immunova AI ML Researcher · Cancer Therapy Models
Iddu AI Founder · Apple TestFlight Beta

Research & Work

From single-cell RNA sequencing to clinical ML platforms, I work on problems at the frontier of computational biology and AI.

2026 – Present
MIT Critical Data
Machine Learning Research Intern
  • Investigate whether multi-agent LLM committees for clinical decision support can be gamed by shortcut cues — signals a benchmark rewards but a clinician would ignore — across seven cohorts on six public datasets spanning text (MedQA-USMLE, MedMCQA, MIMIC-CXR), imaging (NIH ChestX-ray14, MIMIC-CXR-JPG, CheXpert), and tabular ICU records (SUPPORT2)
  • Show that agents resist isolated cues (5–16% answer flips) but that socially plausible shortcuts spread: two peers asserting the same wrong answer, or a false "pre-screen" system flag, flip the holdout agent in 38% of cases across both capability tiers
  • Benchmark three oversight designs — a gate (100% false-positive rate), a same-lineage transcript judge (100% precision, 93% recall on text but collapsing on imaging), and a referee that privately re-queries the holdout (77–88% precision, 13–21% false-positive rate) — establishing that only independent re-querying transfers across modalities
  • Demonstrate that rubric gaming is near-silent in self-report: only 1/10 text and 1/134 imaging drifters name the rubric they moved toward
LLM AgentsMulti-Agent SystemsClinical AIPythonBenchmark Evaluation
2026 – Present
University of Southern California
Quantitative Pharmacology Research Intern
  • Develop mathematical pharmacokinetic models and an agentic AI pipeline to predict the PK behavior of bispecific antibodies, building a structured model repository grounded in their physicochemical properties and assessing structural identifiability against consensus in the literature
  • Extract PK data from clinical papers to support a framework for scaling animal testing results to human predictions based on antibody characteristics, with modeling advanced in parallel with data extraction
PharmacokineticsBispecific AntibodiesAgentic AIRAGMathematical Modeling
2025 – Present
Georgia College & State University
Computational Biology Research Intern
  • Apply machine learning and statistical modeling to identify waterborne pathogens and virulence
Machine LearningStatistical ModelingPythonComputational Biology
2025 – Present
Iddu AI iddu.app ↗
Founder & Lead Developer
Apple TestFlight Beta
  • Founded and independently built an AI-powered Wellness Assistant that autonomously contacts wellness offices via phone to schedule, reschedule, and cancel appointments on behalf of users
  • Engineered the system to navigate complex automated phone trees and IVR systems, reducing patient friction and improving access to mental health and wellness services
  • Designed an end-to-end AI workflow integrating voice automation, NLP, and scheduling logic into a native mobile application
Voice AILLM APINLPiOSHealthcare Automation
2025 – Present
MealsByGrace Club
Co-Founder
  • Co-founded a student-led service organization partnering with MealsByGrace Non-profit to address food insecurity across the North Atlanta area
  • Scaled membership to 150+ students and raised $3,000 in community funding within the first year
  • Organized 13 recurring volunteer and community outreach initiatives across 2025–2026
LeadershipCommunity ServiceNonprofit

Selected Projects

Published and presented research at competitive national symposia.

2026

Shortcut Contagion in Multi-Agent Clinical LLM Committees

Testing whether committees of language-model agents deliberating on a shared workspace for clinical decision support can be gamed by shortcuts — cues a benchmark rewards but a clinician would ignore. Across seven cohorts on six public datasets spanning text, chest imaging, and tabular ICU records, agents resist cues in isolation (5–16% flips), but socially plausible shortcuts spread: two peers asserting the same wrong answer, or a false "pre-screen" system flag, flip the holdout agent 38% of the time. Comparing three oversight designs shows a gate cannot separate adoption from honest agreement, a transcript-reading judge works on text but collapses on imaging, and only a referee that privately re-queries the holdout transfers across modalities — because the drift is near-silent in self-report.

2025

Microparticles as Biomarkers for Blood Coagulation in Cancer

Conducted a comparative analysis of novel microparticle-based biomarkers versus established coagulation markers in oncology contexts. Presented at the IJSCAR Symposium through a competitive selection process.

2025

A Framework for Auditing the Consequences of CRISPR

Led a three-member interdisciplinary research team examining ethical, societal, and regulatory risks of CRISPR gene-editing technologies. Produced a structured audit framework applicable to emerging biotechnology governance and public policy.

What I Work With

Languages & Tools

Python JavaScript SQL Git

AI / ML

Scikit-learn Pandas NumPy Survival Analysis Prognosis Modeling scRNA-Seq Analysis LLM API Integration Voice AI Automation

Domains

Computational Biology Cancer Research Clinical AI Healthcare Automation End-to-End AI Product Development

Awards & Distinctions

Top 3
HOSA State Conference 2026 Medical Law & Ethics Qualified for International Conference · Indianapolis, June 2026
Top 7
HOSA State Leadership Conference Public Health Competition
Top 5
DECA State Conference 2025 Sports & Entertainment Marketing Qualified for ICDC
~Top 25%
IJSCAR Symposium Selected to Present Cancer Biomarker Research Competitive applicant selection

Service & Mentorship

Learn To Be — Nonprofit Tutor

110+ hours · 6 students · foundational literacy & numeracy gains

2024 – Present
Schoolhouse.world — SAT Math Tutor

24+ hour-long sessions · students from diverse global backgrounds

2024 – Present

Let's Connect

I'm always open to research collaborations, internship opportunities, or just a conversation about AI and biotech.