2. BALAJI SHARMA balajirsharma@gmail.com
MS, Mechanical Engineering
THESIS ON VEHICLE VIBRATIONS
Signal Processing, Modal Analysis,
DAQ from Accelerometers, Load Cells
CAREER SUMMARY
1980s 2006 2010 2013 2018
hello world datestr(now)PhD, Mechanical Engineering
STRATEGIC TECHNICAL ENGAGEMENTS WITH HIGHER ED
Supporting Pedagogical Innovations, Learning Initiatives,
Project-Based Learning, Conferences and Workshops
BE, Mechanical Engineering
Wrote my first
piece of code
(only to realize
that world
domination
needed other
skills)
Developed my
first website
(using Geocities,
back when Yahoo!
was cool)
Joined Structural
Dynamics Research
Lab, UCincinnati
(and spun clean,
harmless puns on
shakers, vibrations
and excitations)
Discovered the joys of working
with large public datasets
(sifting through BTS flight data
to prove that Buffalo was better
than Albany, at least when it
came to handling the weather)
Joined Cooperative Distributed
Systems Lab, UCincinnati
(and discovered that robots make
for good conversationalists, despite
the occasional tantrums)
Authored my first
research paper
(and learnt a new
language – obfuscation
academic jargon)
Selected for the NSF I-Corps Program
as Entrepreneurial Lead
(bravely stepped outside the academic
cocoon into the real-world for
research commercialization
and market exploration)
Joined MathWorks (makers of MATLAB)
The mothership called me home
Exploring personal location history for
insights and trends
(Discovering patterns in my movements)
DISSERTATION ON DISTRIBUTED CONTROL
Multi-agent systems, Control Theory,
Ground and Aerial robotic platforms, Motion Tracking
Seeking opportunities in
location intelligence and data analytics
for consumer and market insights
3. BALAJI SHARMA balajirsharma@gmail.com
SKILLS
AND WORKING ON THE REST…
DATA ANALYSIS AND VISUALIZATION
MATLAB ● ● ●
MS Excel ● ● ●
Python ●
Tableau ●
PowerBI ●
WEB AND DATABASES
HTML ● ● ●
CSS ● ● ●
SQL ●
WordPress ● ●
Grav Flatfile CMS ● ●
SENSORS AND PROTOCOLS
Accelerometers ● ● ●
Load Cells ● ● ●
GPS Sensors ● ●
Serial/RS232 ● ●
TCP/IP/UDP ● ●
NMEA/MAVLink ● ●
4. BALAJI SHARMA balajirsharma@gmail.com
DATA PROJECT (WIP): CAN THE WHERE HELP DISCOVER THE WHO?
DATA SOURCE
• Personal Location History from Google
(Timestamps, Latitude and Longitude markers)
• ~235k rows of data over three years
DATA LIMITATIONS
• Sparseness of location data
(spatially and temporally)
• Data/sensor accuracy
OBJECTIVES
• Can anonymized data give away home/work info easily?
• How accurately can place APIs infer venues from raw [Lat, Lon]?
• How well do movements in the real world reveal personal habits?
SAMPLE LOCATION MARKER
EXTRACTED LOCATION DATA
RAW LOCATION MARKERS OVER A THREE-YEAR PERIOD
5. BALAJI SHARMA balajirsharma@gmail.com
DATA PROJECT (WIP): CAN THE WHERE HELP DISCOVER THE WHO?
Despite frequent changes in 2017,
the data gave my home away fairly
easily, even without advanced
maneuvers. I’d suck at hide-and-seek.
I’ve spent much time
in a region where
Tacos are Mighty.
What is the life event, you ask? Define active, you say?
6. BALAJI SHARMA balajirsharma@gmail.com
DATA PROJECT (WIP): CAN THE WHERE HELP DISCOVER THE WHO?
ASSUMPTIONS
• This analysis relies solely on personal
location data from Google over a three-
year period.
• The focus is largely on time spent at
venues (businesses, residence, work).
Location data associated with
commute/drive are ignored for the
preliminary analysis.
• Place lookup APIs (Google, Foursquare)
are used to identify the most likely
venue within ~100 ft of the [Lat, Lon].
METHODOLOGY
Preprocessing
• Raw location data is initially preprocessed and filtered for the desired range of years.
• The accuracy of the location markers across the dataset is evaluated over a quick histogram check, with the
options to use the entire dataset or work with a filtered one.
• All points are visualized on a global map to identify any obvious anomalies.
Clustering
• The points for each day are clustered over a short spatial distance using a density-based clustering algorithm
(DBSCAN) to eliminate all drive and flight markers, allowing focus on visits to venues. The centroid for each
cluster, and the earliest and last timestamp for each cluster, are extracted to estimate dwell times as each venue.
Home Location Estimation
• Clusters for each day are analyzed as monthly aggregates, and as a preliminary approach, a weighted ranking
scheme is employed to identify home locations for each month based on cluster size of markers for each
location, the temporal spread of the markers for each month and estimated dwell time at each location
• Other heuristics in consideration are (reference) the ‘last’ Destination each day and the largest clusters for
specific times of the day.
Venue Estimation
• Foursquare Places API and, for comparison, Google Places API are employed to match [Lat, Lon] to venues and
businesses within a 100 ft distance, and the estimated venues are manually checked for accuracy and identifying
dwell times, behaviors and insights (temporal patterns, preferences for restaurants, departmental stores, etc.)
DETAILS
Data Source
Google Location History (JSON)
Place Lookup
Foursquare Places API, Google Places API
Analysis and Visualization
MATLAB, Python (Folium/Leaflet),
Excel, Google Static Maps API
7. BALAJI SHARMA balajirsharma@gmail.com
DATA PROJECT: HOW DO AIRPORTS FARE AT ON-TIME PERFORMANCE?
EXPLORATION OBJECTIVE
Would a move from Albany to Buffalo
impede my ability to travel by air, given the
harsher weather at BUF?
Would the move cost more in airfare
expenses?
DATA SOURCE
Bureau of Transportation Statistics -
Flight data for 2011-2015
(~30M rows of data)
OUTCOME
Buffalo, for a larger airport than Albany, had
a comparable on-time performance vis-à-
vis weather-related delays and
cancellations, and a much lower average
cost of flight as well.
PS. The assault on colors and aesthetics, in hindsight, could have been avoided.
8. BALAJI SHARMA balajirsharma@gmail.com
IN A LIGHTER VEIN
I once took the train(s) across the US, coast to coast and back, to see how much
of the US I could see without stepping into an airplane or a car.
The journey was great, and after a week on wheels, even the frozen burgers on
the trains started tasting good.
The coffee, sadly, didn’t.
Much of my grad research happened in a lab, fondly called the ‘Dungeon’
(the ‘Cave’ was taken). three levels below human habitation.
The vending machine could only spit out so much coffee past midnight,
forcing us to find other sources of caffeine to fuel the research engine.
For science!
Speaking of locations… Paying my Dews…