You are on page 1of 2

projects

POI Hours Conflation:

The most challenging task that I have worked on is POI Hours Conflation. In Maps
ecosystem, POI refers to Point Of Interest.for eg (Restaurants, Malls, Hotels).
For each attribute in a place eg: name, url, phone, hours, presence
period(open/close) Conflation job is to pick which VendorPlace provides the best
attribute for the place and pick the attribute from those VendorPlaces.

How:
calculate weights for the hours sources based on the business logic
implement tie breaker rules

key words(vendor-country-mapping, brands, prioritized vendors, edits)


Sources for hours: Web scrapping, Vendors, Edits, Try Rating hours
Challenging part:
Had to deal with large dataset and perform map reduce operations using spark
in an optimal way.
Write the rules for rule based model in an optimal and a particular way.
Rules based and ML model. ML model predicts the scores for the sources
I had to explore deep into the Maps architecture to understand the POI
properties

vendor ingest dashboard


-- multiple vendors provide POI data to Apple Maps. (Yelp, zomato, mapmyindia)
-- flow is like vedor uploads the data to amazon s3 servers and that data will be
downloaded and stored in Apollo(Internal service) and it undergoes some approval
and conflation process and finally data is ingest in maps
-- pain points:
- no visibility on when to expect next delivery
- mismatch between vendor delivery and apollo delivery status
- identify missed deliveries
- noise from inactive vendors

features:
showing failed steps
showing cadence
graphical representation
jenkins --- local db store
historical trends page

Process Anomaly Scoring:

-- feature data fix


===================================================================================
===
POI Augmentation
Process Anamoly Scoring
-- provide scoring for anomolies i.e to the POI groups so that we can
assign some priority to them whenever we want to perform some action on that
Hours Conflation
--
NSP Web Application
Vendor Ingest dashboard
-- Aim is to identify the missing deliveries from different vendors and
notify them or report them.
-- automated several sequential tasks(hercules jobs/pipelines) which
reduced the manual effort by 100% and the wait time between two tasks came down to
zero.
- Analysed the historical data of different jobs and allocated the
memory resources accordingly
Feature Data Fix
-- created a CLI and Ui for fixing the feature data in bulk.
-- complex json structures
-- own thought process

Automated mail

Neutron CLI
CLI for executing business tasks, which include authorizing the users,
providing access, filtering queries, job run queries, project queries etc

Video Sequence Prediction


Automatic Rubik's Cube Solver

ReactorX Jobs execution

Automation - Kafka jobs for even handelling

You might also like