- OUR MISSION & STORY
We strive to help our clients derive insights from unstructured data using machine learning & curation.
In naming our company, we drew inspiration from a captivating story from the ancient Indian epic Ramayana. As the story goes, Lord Rama oversees the construction of a gigantic bridge, called Setu in Sanskrit, between India & Sri Lanka in order to cross the ocean & rescue his queen.
The phenomenal task is accomplished by recruiting a large army of volunteers & through divine powers that help the bridge stay afloat!
This story is strikingly allegorical to our company’s goal, namely, helping our clients cross the “ocean” of big-data using our services which act as the bridge between unstructured & structured data! In order to accomplish this goal, we use a combination of vast human intelligence (our associates) & mystical powers (Machine Learning) as well! Thus was born the name SetuServ, from “Setu” & “Services”.
- FOUNDING TEAM
Sanga Reddy Peerreddy
Founder & CEOSolves big data problems, passionate about social enterprisesZS, IIT Mumbai, Purdue, Kellogg
Founder & CMOFocussed on practical solutions to complex problemsZS, Tech Mahindra, UIC, Booth
FounderNLP Guru, Thought leader on combination of machine learning with human computation
Stanford, CMU, IIT Mumbai
Prabhakant SinhaFounder of ZS Associates and Professor at Kellogg School of Management
D SahayManaging Principal, Business Technology Lead, ZS Associates
Harsha focuses on value-oriented investing. His prior professional experience includes private equity investing, angel investing, strategy consulting (McKinsey & Company), and data analytics. He has worked in the US and in India.
David VincaFounder of eSpark Learning and successful entrepreneur
Arjun AggarwalCo-founder Healthscape Advisors
- OUR RESEARCH
Tech Talk (Delivered on May 13, 2013)
SkierarchyTM approach and it’s applications.
Google Tech Talk (Delivered on Feb 20, 2013)
Can Crowdsourcing Replace Experts? How We Attained The Quality of NIST’s Analysts Using a Hierarchy of Experts, Crowdsourcing and Machine Learning at the TREC Crowdsourcing Evaluation.
As unstructured Big Data in the form of textual, image, audio and video content has continued to grow exponentially in recent times, the need to derive insights from this data has emerged as a significant scientific and technical challenge. While advances are being made in the field of Machine Learning (M/L) to address this issue, current M/L techniques are yet to reach human-level performance on many domains such as natural language, speech, image and video interpretation. As a result, the idea of crowdsourcing, that deploys human intelligence at scale, has been gaining currency as an alternative to Machine Learning.
Current crowdsourcing approaches are successful for tasks that are simple enough for untrained people, but are yet to be demonstrated to be appropriate for more complex tasks that require domain-specific skills. Extending crowdsourcing to such skilled tasks at scale requires overcoming several challenges including (but not limited to) the following:
- Skilled tasks require domain-specific skills or expertise, which an ad-hoc crowd may not possess.
- Challenges in solving skilled tasks are not always evident apriori, making it hard to develop comprehensive training or guidelines upfront.
- Complex tasks that are difficult for an annotator tend to be difficult for others as well, making the popular solution of redundant annotations relatively ineffective.
- Desired outputs for complex tasks are not known upfront, making it hard to measure and control for quality.
- The cost of crowdsourcing increases linearly with data size, making it unsustainable at scale.
We address these challenges using a novel hierarchical approach involving a small number of domain experts at the top of the hierarchy, a large crowd with generic skills at the intermediate level, and a Machine Learning system serving as a personal assistant to the crowd, at the bottom level. We call this approach SkierarchyTM, short for Hierarchy of Skills.
To test the efficacy of the SkierarchyTM approach, we deployed the model on the TREC 2012 TRAT (Text Relevance Assessment Task), a task that is fairly more complex compared to typical crowdsourcing tasks. In this talk, we will present experiments to demonstrate the utility of each of the layers of our hierarchy in terms of meeting the challenges mentioned above. Our experiments on TRAT show that our approach attains performance levels on par with those of NIST’s expert assessors.
We successfully demonstrated our SkierarchyTM approach at the Text Retreival Conference (TREC) crowdsourcing track. This conference was organized by National Institute of Standards and Technology (NIST). A key objective of this reputed academic conference was to encourage research into crowdsouricng techniques for obtaining high quality relevance judgments for document-topic pairs. Our relevance annotations were found to be highly accurate as evaluated with respect to those of NIST’s own professional assessors. For additional details, please read our TREC paper published at TREC Proceedings.
222 Merchandise Mart Plaza, Suite 1212
Chicago, IL 60654
Orange 108, My Home Rainbow,
Shaikpet, Hyderabad, AP, India 500008