Birla Institute of Technology and Science, Pilani

Software Engineer at udaan.com

Skills: Project Management · People Management · Leadership · Agile Methodologies · System Design · Team Building · Stakeholder Management

Flipkart

Tech Lead Manager

- Owning Manual Annotations charter serving ~5 million image based per month and ~2 million video based annotations per month, for ML models training and evaluation.
- Owning offline ML batch processing platform used by 10-15 internal teams, with ~60K batch jobs per month for ML models data-prep, evaluation and experimentation.
- Leading a team of 7+ SDEs working backwards from continuous business goals

Amazon

Software Development Manager

- Built Kibana based near real-time dashboard to provide transparency of manual annotations (labeling tasks) throughput and backlog.
- 1-click submission of annotations using workflows by automating manual steps, reducing TAT from ~8 hours to <2 hours.

Software Development Engineer 2

- Revamped logistics charges frameworks based on Geo, client(Buyer/Seller), and business category pivots from scratch.
- Mentoring two junior software developers with the cognizance of product asks vs tech implications, code quality via code reviews, futuristic design (HLD/LLD) considerations, and soft skills.

udaan.com

Software Engineer

~ Non Functional Requirements (NFR) benchmarking/optimisations for Search Federator and Conversational Search, Big Billion Days 2018.
          # Scaled Search Federator to 100K QPS with P99 latency of 550 ms with load average of around 0.9.
          # Scaled Conversational Search to 65K QPS with P99 latency of 55 ms.
          # Deep dive and fine tuning on the number of threads in the Hystrix Pool, and analysing the number of context switches happening at OS level. Coming up with optimum amount of threads per upstream network calls.
          # Introduced request cache in Conversational Search, leading to 5X drop in P99 latencies and 1.5 X increase in throughput, per server.
Technology & Tools used : Java, Grafana, Hystrix(Netflix), CouchBase, Locust, Guava Cache, Dropwizard etc.

~ Built Conversational Search 2.0 from scratch.
          # Ownership of the HLD and the LLD of the entire pipeline.
          # Highly scalable unbounded candidate generation of inorganic search contents. ( 4X coverage)
          # Fully transparent and ease of debuggability : What is getting generated and Why?
Technology & Tools used : Java, Cascading(Java) MR Jobs, MongoDB and CouchBase.

- Design and implementation of a new (C++ based)stack(from scratch) for TurboTax Desktop. This new stack saves the extra maintenance cost of the legacy stack, and opens up more paths towards innovation and customer delight.
- Integration of the above stack with TurboTax Windows (C#).
- Integration of the above stack with TurboTax Mac(Objective –C).

Intuit India

- Conducting an A/B experiment on 5 million TurboTax customers with different promotional offers. The aim of the experiment was to come up with the best offer, to enhance customer retention and new customer acquisition. 
 - Enhancing the TurboTax installer success ratio by fixing filesystem permissions related issues. Resulted in the reduction of ~4K customer care calls.

Software Development Engineer 1

 Profiling and Logging using AspectJ in order to test and enhance code productivity.
 Evaluating different graph databases like Neo4J, OrientDB, Titan+HBase etc. on the basis of
import and traversal time using Tinkerpop APIs.

Tushar Bihany