Jagadish Munagapati

Georgia Tech, CSClass of 2029

I build full-stack apps and research software. I'm looking for a software engineering internship for Summer 2027, in backend, full-stack or machine learning work.

JM-01FOMO, a social calendar that's live on the web and the App Store.
JM-02, 03Two quant research projects that charge every trading cost.
JM-05Index Edge: 78 pre-registered SPY strategies. None beat buy-and-hold.
JM-04A heart-risk estimator that reads a facial photo.
PUBSix medical research talks, posters and abstracts, 2025.

Email me GitHub

Windows drag by their title bar and resize from the corner. Open more from the icons on the left, or type help in Terminal. Esc closes the front window.

FOMO

Social calendarWeb and iOS

Your status comes from your calendar, and friends see it change live. I built it for making plans without the group chat.

Open fadethefomo.com Download on the App Store

FOMO iPhone screen: Friends on a map, each with a live status.

FOMO iPhone screen: a friend's card with status and location sharing.

FOMO iPhone screen: The O, a timed status override dial.

FOMO iPhone screen: a group and its members.

FOMO iPhone screen: group events with RSVPs.

FriendsFriends on a map, nearest first, each with a live status.

Features

Live status: busy, free or asleep, from your calendar.
"The O": a timed manual override.
Real time over Redis pub/sub and one WebSocket.
Two-way sync with Google Calendar and Tasks.
Schedule import from a photo, read on the device.
Groups with events, RSVPs and invite links.

Specifications

Role	Developer: backend, web and iOS
Status	Live on the web and the App Store
Backend	FastAPI, PostgreSQL, Redis, WebSockets
Web client	React, MapLibre
iOS client	SwiftUI, MapKit, WidgetKit, APNs
Source	Private, available to walk through

Figure 1.Status logic, interactive. Drag across the calendars or press play.

2:30 PM

Drag across the calendars to move through the day. Each status comes from that person's events and sleep schedule.

Same rules as the real app, with sample people. In FOMO the statuses update live over a WebSocket whenever someone's calendar changes.

Crude oil microstructure

Quant researchWTI crude futures

Forecasts the next 60 seconds of crude oil futures from order flow, then trades the forecast with every cost charged.

Key results

Parameter	Value	Conditions
Net profit	+$6,390	30 walk-forward test days, all costs
Holdout net profit	+$3,510	10 days, run once
Placebo p-value	0.005	vs 200 shuffled runs
Break-even fee	$6.35	per round trip
Deflated Sharpe	0.20	after 46 configurations tried

Figure 1.Cumulative profit before and after costs. Hover for each day.

After all costsBefore costs

Spread, fees and slippage take about 70% of the gross profit.

Figure 2.Placebo test, 200 shuffled runs.

Figure 3.Sensitivity to costs.

After costsBefore costs

Figure 4.Most predictive features by forecast horizon.

Features

78 features, each tested for lookahead.
Purged walk-forward training, no leakage.
Realistic fills: bid/ask, book depth, fees.
11 risk checks on every order.
Placebo and cost tests on every result.

Limitations

Not evidence of a tradable edge: the data is simulated, and the model recovers a signal planted in it.
Trying 46 configurations inflates the best result; the deflated Sharpe ratio is 0.20.
What it does show is a full research pipeline that guards against lookahead, leakage and unrealistic fills.

Specifications

Models	OLS, Ridge, Lasso
Data	Calibrated market simulator, frozen before research
Stack	Python, pandas, NumPy, scikit-learn, statsmodels, pytest
Tests	GitHub Actions CI

Code and full report

Momentum Trader

Quant researchUS small caps

A pre-registered test of a popular retail small-cap momentum strategy. After real trading costs, it did not make money.

Figure 1.Average result per trade, in R (1R is the amount risked).

Before costsAfter real quote costsRandom entries, same costs

It beats random entries in every group, but after real costs it is flat or negative in every year, price range and gap size: −0.14R per trade overall.

Features

Gapper scanner with ten safety filters.
Ten setup detectors, each proven lookahead-free.
SEC filing checks for offerings and reverse splits.
Random-entry baseline on every test.
Pass/fail rules written before the test.

Specifications

Data	About 4 years of 1-minute bars; real Nasdaq quotes for 893 trades
Quote data cost	$15.23
Stack	Python, pandas, Alpaca market data, Databento, SEC EDGAR
Money at risk	None; no broker was connected

Price chart of XPON on August 14, 2025, marking the entry above the premarket high. — Figure 2.One simulated trade: XPON, August 14, 2025. Entry when price first closed above the premarket high.

Code and findings

CVD Risk Estimator

Machine learningHealth research

Estimates 10-year heart disease risk from the ASCVD equation plus four classifiers that screen a facial photo for warning signs.

Live app: cvd-risk-estimator-jagadish.streamlit.app

Open the app View the code

Features

Four image classifiers on DeepFace embeddings.
Screens for xanthelasma, neck veins, age spots and neck size.
ASCVD score from clinical inputs.
Model comparison with cross-validation.

Specifications

Presented	Regional Science & Engineering Fair; Sigma Xi Research Online Symposium
Stack	Python, scikit-learn, DeepFace (VGG-Face), Streamlit
Since	August 2024
Live app	cvd-risk-estimator-jagadish.streamlit.app
Note	For education and research only, not medical advice

The CVD Risk Estimator input form. — Figure 1.The input form. The hosted app sleeps when idle, so the first load can be slow.

Index Edge

Quant researchSPY, 1993 to 2024

Can a simple, rule-based SPY strategy beat buy-and-hold after realistic costs? I tested 78 pre-registered settings across ten strategy families. None passed.

Code and results Final report

Key results

Parameter	Value	Conditions
Settings tested	78	10 families, each written down before any code ran
Strategies that passed	0	all 8 checks required, out-of-sample
Closest: pre-FOMC drift	6 of 8	+0.35% per trade, 8% max drawdown, β 0.08
Bad vendor prices found	176 days	Alpaca daily closes vs official closes, 2016 to 2022
2025+ holdout	Unused	kept for a candidate that passes; none did

Figure 1.Growth out-of-sample, after stress costs. Pick a family; hover for each month.

Figure 2.The pass rule. A strategy needed all eight.

Family	Trades	Per trade	1	2	3	4	5	6	7	8

200+ trades
Profitable at stress costs, profit factor 1.2+
70%+ positive years, none below −10%
Max drawdown 20% or less
Beats 95% of random-timing runs
Edge left after removing market exposure
Deflated Sharpe 0.95+ across all 78 trials
Same settings profitable on QQQ and IWM

Figure 3.Return by year, stress costs.

Up yearDown year

Figure 4.Pre-FOMC drift in bad markets.

Pre-FOMC driftSPY buy-and-hold

Holding SPY into scheduled Fed announcements made money while SPY fell, but it was in the market only 6% of the time.

Features

Rules and settings pre-registered before any code ran.
Walk-forward: pick on 24 months, test the next 6.
1¢ and 3¢ a share costs, plus SEC and FINRA fees.
50 random-timing runs and buy-and-hold on every result.
Deflated Sharpe penalty for all 78 trials.
Checked on QQQ and IWM, never tuned on them.

Findings

Intraday SPY edges were a cent or two a share, smaller than costs.
Daily edges came and went with the market, and were mostly market exposure.
Alpaca had SPY at 255.24 on 2020-03-12; the official close was 248.11.

Specifications

Data	SPY 1-minute bars 2016 to 2024; daily bars 1993 to 2024, checked against official closing prices; Fed meeting calendar 1994 to 2024
Stack	Python, pandas, pandas_market_calendars, Alpaca, Yahoo, Databento, pytest (42 tests)
Data cost	$0.007 of Databento credit
Money at risk	None; no orders were ever placed

Research

Six items, 2025Talks, posters and abstracts

Medical research I co-authored before college, mostly on national health datasets such as CDC WONDER and NEISS.

Date	Title and authors	Venue
Nov 2025	Seasonal Variation in Consumer Product–Related Burn Injuries in the United States: A National Survey AnalysisMunagapati S., Goyal A., Munagapati J., Kapur S.	Podium presentation, Northeast Region Burn Conference, Washington, D.C.
Nov 2025	Trends in Burn Injuries from Consumer Products: A Decade-Long Analysis of NEISS Data (2015–2024)Goyal A., Munagapati S., Munagapati J., Kapur S.	Poster, Northeast Region Burn Conference, Washington, D.C.
Nov 2025	Trends in Brain Tumor Mortality in the U.S. by Race and Sex, 1999–2020: A Population-Based AnalysisMunagapati S., Goyal A., Kapur S., Munagapati J.	Abstract accepted to Neuro-Oncology
Nov 2025	Temporal and Demographic Trends in Brain Tumor-Associated Emergency Department Visits in the United States: A CDC WONDER-Based AnalysisMunagapati S., Goyal A., Kapur S., Munagapati J.	Abstract accepted to Neuro-Oncology
Sep 2025	Shared Risk, Stronger Care: Risk Pooling for Rural SurgeryMunagapati J., Bazal C., Coule L.	Poster, Georgia Rural Health Association Conference, St. Simons Island, GA
Jul 2025	Real-Time AR-Guided and AI-Optimized Port Placement for Robotic and Laparoscopic SurgeryMunagapati S., Goyal A., Munagapati J., Kapur S.	Oral presentation, Society of Robotic Surgery, Strasbourg, France

Other projects

Three itemsTeam and volunteer work

Project	Description	Links
Student Finance Manager	Budgeting app built with a teammate for FBLA. Charts spending, reads a photo of a paper check, and answers questions about your own data.	Live app Code
WolfRow	Rowing trainer, built with a collaborator. Logs workouts, flags fatigue from pace trends, and plans training weeks.	Code
Augusta Free Surgery Clinic website	Responsive site and outreach graphics for a nonprofit that provides free surgery. Volunteer work since 2024.

Experience

Georgia Tech, CSClass of 2029

Dates	Role
Aug 2026 to present	Software Engineer Trainee, Georgia Tech Molecular Evolution CoreFront-end and back-end work on three active projects in the lab's codebase.
Oct 2024 to present	Marketing Advisor and Web Developer, Augusta Free Surgery ClinicBuilt the clinic's website; runs its digital and print outreach.
Jan 2023 to present	Chapter President, Back on Track America (Georgia)Leads a 70-member volunteer chapter that has saved partner organizations over $50,000.

Contact

Email is the best way to reach me, at my Georgia Tech address first; my personal address works too. I'm open to SWE internships for Summer 2027.

jmunagapati3@gatech.edu jmunagapati@gmail.com GitHub LinkedIn

jm@portfolio ~ %

Jagadish Munagapati, portfolio