Currently building Aro

Soumil
Bhandari

0
Patents Filed
0+
Active Users
$0M+
Deals Surfaced
Soumil Bhandari

Researched at · Built for · Trusted by

Harvard Medical School
Mass General Hospital
Mass General Brigham
MIT
IIT Roorkee
McKesson
Tekizma
Stanford University
Columbia
Dartmouth
ASTM International
Harvard Medical School
Mass General Hospital
Mass General Brigham
MIT
IIT Roorkee
McKesson
Tekizma
Stanford University
Columbia
Dartmouth
ASTM International

Projects

Shipped & Built

Real products. Real users. Research that earns patents.

StealthNov 2025 – Present

Stealth Startup

Modernizing Neuroscience.

Details under embargo. More to come.

In Development2025 – Present

NeuraBand

A sub-$80 wearable coming to market soon — catching Alzheimer's early enough to slow it down.

Continuous passive monitoring of 10 neurologically relevant biomarkers, streamed over BLE to a clinical-grade dashboard. Designed to surface prodromal neurological signals early — when intervention can still slow progression. Costs under $80 to build. Coming to market soon.

0
Biomarkers
<$0
Cost
FirmwareMAX30102BMI270 IMUBLEChrome Dashboard
Live2025

Flipper

500+ flip-ready properties surfaced. $1.2M+ in projected profit.

Ingests Bright MLS data, geocodes every listing, and scores each one on comp-based ARV, rules-based rehab estimates, DOM percentile, and tax-assessed discount — outputting a verdict of BUY_NOW, LOWBALL, VERIFY_FIRST, or AVOID. An optional AI-assisted layer brings Claude in as a second underwriter on every deal — explicit ARV ranges, rehab estimates, max-offer, and risk flags — for a fraction of a cent per listing thanks to prompt caching. Has surfaced 500+ flip-suitable properties with $1.2M+ in projected profit. Personally made $50K+ on flips it found — projected to clear multiple 6 figures by end of 2026.

0+
Properties Found
$0M+
Projected Profit
PythonPostgreSQLPostGISClaude AIBright MLS
LiveApr 2026 – Present

Standerds

A new file format for documents that think.

Every document format since PDF is just styled text. Standerds introduces a 2D hierarchical format for standards, reports, manuals, and policies — built so you can navigate, search, and synthesize content the way it was actually meant to be read. In the process of becoming an ASTM reseller. Coming to jobsites very soon.

Live
Status
ASTM (in progress)
Distribution
ReactNode.jsMongoDBNLPData Modeling
Visit standerds.com
LiveApr 2024 – Mar 2026

Notes360.ai

AI note-taking, used by students across five schools worldwide.

Built end-to-end: responsive cross-platform UI, scalable backend, AI-powered note processing, and weekly feedback loops baked into the product cycle. Grew to 1,000+ active students across 5 school partnerships. Full lifecycle from zero to shipped.

0+
Users
0
Schools
React NativeNode.jsMongoDBGoogle CloudHugging Face
Live2024

Rhetorification

9,000+ uses. The evidence engine that ran an entire debate team's prep.

Lightweight Python utility that parses Microsoft Word (.docx) files, extracts highlighted and bolded citation passages along with their preceding context tags, and renders each snippet as a clean, browser-viewable HTML card. Purpose-built for competitive Public Forum evidence prep — drag in a doc, get back a navigable card stack. Adopted by an entire school debate team and counting 9,000+ uses.

0+
Total Uses
Full Debate Team
Adoption
Pythonpython-docxFlaskJinja2BeautifulSoupHTML/CSS
Visit rhetorification.online
Built2026

Locked In Meter

Real-time focus tracking — no wearables, just your webcam.

Multi-modal computer vision pipeline that tracks eye movement, head position, and hand motion simultaneously through your webcam, synthesizing all three signals into a live engagement score. No EEG, no wearables, no self-reporting — just a model that tells you whether you're actually locked in or just staring at the screen. Built because I wanted ground truth on my own work sessions.

0 (eye + head + hand)
Tracked Signals
Webcam Only
Hardware
PythonOpenCVMediaPipeNumPyReal-Time CVPose Estimation
ResearchFeb 2026 – Present

ATLAS

ML pipeline for classifying motor evoked potentials from EMG.

Closed-loop TMS motor threshold estimation system. Automates MEP detection with calibrated confidence scoring and uses Bayesian entropy-minimizing adaptive algorithms to find resting motor threshold efficiently. Full signal processing pipeline end-to-end: bandpass + notch filtering, baseline correction, and 20-feature ML classification with LOSO cross-validation. Confidence-weighted updates, real-time data quality monitoring, operator override with audit trail and full trial logging.

0
Features
Synthetic
Data
Pythonscikit-learnSignal ProcessingNumPySciPy
Open Source2026

nanoDLM

The nanoGPT of masked diffusion language models.

The simplest masked-diffusion LM you can actually train, debug, and learn from — ~1000 lines of pure PyTorch, no tokenizer, no diffusers, no HF Trainer. Fixes real bugs in the original MDM objective and sampler, modernises the backbone with RoPE and self-conditioning, and benchmarks MDM head-to-head against a matched autoregressive baseline. Supports infilling — something AR architecturally cannot do.

~0
Lines
Pure PyTorch
Framework
PythonPyTorchRoPEMasked DiffusionLM Research

Experience

Work & Research

From Harvard labs to healthcare giants — always at the frontier.

01
Nov 2025 — Present
Entrepreneurship

Stealth Startup

Modernizing Neuroscience
Founder & CTO

Modernizing neuroscience. Details under embargo.

AI/ML3 PatentsClinical PilotsStealth
02
Feb 2026 — Present
Research

Harvard Medical School

TMS Lab
AI/ML Researcher

Conducting research in the Transcranial Magnetic Stimulation Lab under Dr. Aapo Nummenmaa. Developed ML classification algorithms for electrophysiological signal analysis and motor threshold estimation — achieving the best classification results the lab has seen to date. Results led to a second ongoing project and an incoming publication.

PythonMATLABNeurosciencePublication Incoming
03
Mar 2026 — Present
Startup

Slait

YC-backed
Founding Intern

Improving AI session classification pipeline and LLM-based scoring across 8 engineering dimensions. Driving GTM strategy and user growth for an AI coding analytics platform backed by a YC interview.

LLMGTMYC-backed
04
May 2025 — Jan 2026
Research

IIT Roorkee

India
AI/ML Research Intern

Postgraduate-level research at one of India's top technical institutions — as a high schooler. Built Random Forest and XGBoost classifiers hitting 92%+ accuracy, fine-tuned LLaMA-2 and Flan-T5 via LoRA for specialized NLP tasks, and shipped autonomous ML pipelines on distributed cloud infrastructure. Received awards for outstanding contributions and was personally invited for a campus tour — the only intern.

XGBoostLLaMA-2LoRAOutstanding Contributions Award
05
Apr — Jul 2025
Internship

McKesson

Fortune 10 Healthcare
Software Engineering Intern

Built GPT-powered systems that processed 1M+ unstructured clinical records and cut analytics time by 35% across enterprise workflows — at scale, in production, at one of the most complex healthcare infrastructures in the US. Designed TensorFlow and PyTorch ML pipelines integrated directly into live EHR systems. Shipped backend services in Python and Java that are actively running today.

GPTEHRTensorFlow1M+ Records
06
May 2022 — Apr 2025
Internship

Tekizma Inc.

Enterprise AI
Software Engineer

Built production-grade AI-driven software solutions for enterprise clients still actively referenced by senior engineers today. Developed NLP modules using spaCy and Hugging Face for 10K+ entry datasets, and built full-stack applications (React, Node.js, MongoDB) for real-time data processing at scale. Received the Distinguished Intern Award and Outstanding Contributions Award — the only intern recognized multiple consecutive years.

spaCyReactNode.jsDistinguished Intern Award

Awards

Recognition & Honors

3 patents. 3 championships. 5 TOC bids. The receipts.

🎙️

2024 – 2025 Season

Competitive Debate

Public Forum · Top-Ranked Debater in School History

3 Championships5 TOC Bids3× Gold Bid
ChampionColumbia InvitationalNation's most competitive PF tournament11–1Jan 2025
ChampionWACFL Public Forum 5Undefeated at largest Regional Tournament to date4–0Feb 2025
ChampionWACFL Public Forum 4Outperforming top DC-area teams4–0Jan 2025
QuarterfinalistGeorge Mason University1st overall seed after undefeated prelims · Gold Bid to TOCUndefeated prelimsDec 2024
QuarterfinalistCoolidge Qualifier1st seed in national-level PF qualifier · Gold Bid to TOCUndefeated prelimsJan 2024
OctofinalistSunviteGold Bid to TOC at elite national PF tournament5–1Jan 2025
🔬

2025

Virginia State Science Fair

Leidos · Yale Science & Engineering Association · Office of Naval Research

Alzheimer's progression AI recognized across three independent judging panels — Leidos, Yale S&E Association, and the Office of Naval Research.

🏅

2023, 2024

Distinguished Intern Award

Tekizma Inc. — 2023 & 2024

Only intern recognized multiple consecutive years. Also received Outstanding Contributions Award.

📚

2024, 2025

AP Scholar with Distinction

CollegeBoard

12+ AP classes and 6 Capstone courses. Recognized multiple consecutive years.

Skills

Technical Stack

The full stack — from firmware and fine-tuning to production systems.

Languages
PythonJavaJavaScriptSQLMATLABBash
Frameworks & Libraries
TensorFlowPyTorchReactNode.jsExpressFlaskHugging FacespaCyLangChainLoRA
Developer Tools
GitDockerMongoDBAWSGoogle CloudVercelAgile/ScrumVS Code

Athletics

Beyond the Lab

What happens when the compiler isn't running.

🏀

4

Years Varsity

Varsity Basketball

4-Year Starter — BASIS Independent McLean

Four-year varsity starter — not a single season on the bench. Competed at the highest level of high school basketball from day one.

Varsity

Sophomore Year

Varsity Soccer

Varsity Starter — Sophomore Year

Beat out upperclassmen for a varsity starting spot in sophomore year. The same competitiveness that drives everything else.

Interests

🏀BasketballSoccer🎙️Debate♟️Chess🤖AI Research🚀Building Products

Languages

EnglishNative
HindiFluent

06. Contact

I'd Love to Connect!

Whether you're building something interesting, working on AI or neuroscience, or just want to chat — I'd genuinely love to hear from you. My inbox is always open.

Say Hello 👋

soumil.bhandari@gmail.com

LinkedInGitHub

Designed & Built by Soumil Bhandari