Mar 2022 – Feb 2024 · 1 yr 11 mos · Pune, Maharashtra, India

• As Head of Data Science, driving a global team of 50+ Data Scientists to deliver innovative ML solutions in healthcare. Covering variety of EMR data (diagnoses, procedures, vitals, etc.) like HL7, FHIR, CCDAs. Along with Medical Charts, Claims, Provider directory data to solve following use cases:
• ⇨ Roster Auto-Ingestion system: Uses AI to auto-analyses data and transforms it in the form needed by internal databases. This solution is catering to 100K Providers groups across multiple markets within a national health plan that is among the top-5 in US.
• ⇨ Medical Chart analysis:
• o Chart Completeness: Identifying missing sections in a medical chart.
• o Code detection: Unstructured medical text processing to obtain ICD-10 codes.
• o Ontology based medical topic detection: Using UMLS ontologies to discover code relationships and bring out prominent disease trends.
• ⇨ DRG likelihood prediction using claims:
• o Analysing patient’s historical claims to build code sequence patterns.
• o Building a clinical behavioral profile of each patient
• o DRG likelihood model based upon clinical behavioral patterns of patient.
• o Used to detect abnormal claims that could be suspected upcoding.
• ⇨ Clinical Data Quality: Real time system to auto-ingest clinical EMR data in HL7, CCDA, FHIR, Flat files format and apply ML based auto discovered data quality checks. Currently analysing 500 million records weekly on a national health plan’s network.
• ⇨ Deployed other clinical use cases:
• o Patient class prediction: Predicting patient class: {In, Out, Home, Nursing, ...}
• o Term mappings: Mapping local code systems to standard code systems.
• ⇨ VBC (Value Based Care) misattribution detection: Analysing massive databases (10+ million-member population) along entire attribution process to detect misattribution.
• ⇨ Claims overpayments detection: Built multivariate statistical ML models to analyse claims data in real time to block suspected overpayment by health plan.

Jan 2021 – Feb 2022 · 1 yr 1 mo

• Leading global Applied Data Science Research team with senior data scientists and PhD folks across USA, Canada and India. Ensuring Applied Data Science delivers value to company's customers, for all its analytical solutions and products. This demands a top-down approach where one starts with a business problem of the customer and translates it into actions that can be achieved by Data Science. Essentially tying together 3 components - {Machine Learning, Business context, Engineering viability}
• Using Applied Data Science for following in-house products:
• => Telecom
• > Operations: ML based explainable SEI (Service Experience Index), NEI (Network Experience Index)
• > VoLTE analytics: ML based explainable CEI (Call Experience Index) to replace static MOS
• > Customer Analytics: Subscriber ML-QoE, Churn prediction models, CSAT prediction models
• > 5G NWDAF: Load analytics, LSI (Load Stability Index) for Network Slice, UE Mobility, UE Communications
• => Thales divisions
• > Cyber Security & Fraud detection: Using in-house research of Behavioural Outlier Detection
• > Avionics: Faulty part prediction
• > Ground Transport Rails: Smart Maintenance of devices
• > Manufacturing: Machine fault detection

Jan 2019 – Dec 2020 · 1 yr 11 mos

• Researching and Architecting ML solutions to deliver an end-to-end Applied Machine Learning solution that works over real-world data in the field.
• @ML Research: Paper in international journal and US Patent in Unsupervised ML
• @ML work areas: Unsupervised ML, Connectivity based clustering, Multivariate time series analysis, Forecasting, Sessionisation, Anomaly Detection, Ensemble of Anomalies, Association mining, Temporal clustering, Classification, Confidence bucketing, MLOps, Drift detection, Ontology knowledge graphs, Sematic similarity, Concept extraction
• @Architecture & designing: Developed end to end ML external ML solutions packaged as a product for Telco consumers and
• Did the above in Security Analytics solution for SOCs, Operations Analytics for NOCs in telco domain, Content Analytics for web / marketing domain. Delivered the above by both solving it via Machine Learning research, architecting via right technology and applying to the right business domain.
• Data Science researched solutions developed:
• > Ensemble of Anomalies via expectation scores
• > Behavioural Analytics via data geometry
• > Sessionisation via stochastic periods

Sep 2016 – Dec 2018 · 2 yrs 3 mos · Gurugram, Haryana, India

• As Data Scientist:
• Researching on new Machine Learning approaches in the domain of Cyber Security
• NSBA (Network Service Behaviour Analytics): An unsupervised ML technique to model network service behaviours. This creates statistical Behavioural models of critical network service in an enterprise network. Any flow deviating from historically seen behaviours is raised as an alert. As it catches the abnormal flows, it can catch the “Unknown” threats, malware attacks, bots happening in the network.
• Data Geometry: Constructing topological spaces via semantic similarity between data attributes
• Session discovery by modelling stochastic periods in time series data
• As Data Science Architect:
• Designing and architecting Machine Learning at BigData scale (TBs of data)
• Architecting Ontologies and scaling it to a billion node knowledge graph
• Designing deep learning systems to work on large scale along with reasonable time to train
• Designing real time machine learning predictive systems to prevent possible threats and issues
• Architecting BigData ML technologies like Spark, Graph DBs, Kafka, Redis, HBase, etc.

Apr 2013 – Aug 2016 · 3 yrs 4 mos · Gurugram, Haryana, India

• As performance architect re-designed many BigData architectures for Hadoop, Yarn and Spark. We were able to increase the performance of the whole cluster by 4 times, from 25 thousand records / sec to 100 thousand records / sec for every node of the cluster.
• We designed in-memory cumulative combiners, a lock free multi-mapper, a hardware cache (L1, L2, L3) aware mapper design, optimal I/O parallelism via Hyper Threading.
• Then we optimized implementations of various algorithms - Murmur Hash, Bloom Filters, PCSA, K-Means, MPH
• We also created many in-house tools to diagnose performance issues over distributed Big Data clusters over Hadoop, Yarn and Spark.
• We developed tools to monitor performance across the complete stack starting from
• Hardware - {L1/L2/L3 cache misses, Instruction to miss ratio, etc.},
• OS - {CPU, Context switches, Core scheduling, Memory, NUMA, I/O, Page faults, etc..},
• JVM - {GC detailed stats, promotion criteria, JIT code cache stats, lock stats, TLAB stats, etc..}, Framework level Hadoop/Yarn/Spark - {Counters for input/output/combine as a time series, stats for Read/Process/Write, time analysis}
• and then some custom application level diagnostic stats.

Nov 2011 – Mar 2013 · 1 yr 4 mos · Noida, Uttar Pradesh, India

• Architect for building an Mobile Video Bandwidth Optimiser for Telco operators. Designed a high performance network solution to optimise 10 gbps data lines. Along with it we were building an analytics engine to monetize data traffic flowing through the telco pipe.

Sep 2010 – Nov 2011 · 1 yr 2 mos

• Turiya Technologies builds an ultra low latency (first of its kind) performance monitoring tool for Java and Native systems, which can do performance diagnosis for mission critical real time applications like the ones used in Financial Industries.
• The tool adds a new dimension to performance analysis, being backed by a strong analytics engine it can do smart fishing underneath the heavy gush of data generated in the information waters of modern software systems to present what is relevant to the user as per his need.

Jan 2009 – Jul 2010 · 1 yr 6 mos

• Lead the team whose objective was to R&D into performance enhancements ways both by developing faster algorithms and looking for new innovations in high performance domain to fasten existing technology. By doing a complete Performance tuning of the whole stack used by a product, spanning across different layers like JVM tuning for Java applications, hardware tuning (HyperThreading, power saving features, etc...) OS level diagnosis, TCP parameters tuning for network / IO throughput then to applications infrastructure tuning like process's core bindings, etc., I used to recommend the developers of respective products to evolve methods for achieving even higher levels of software performance.

Nov 2008 – Dec 2008 · 1 mo

• This US based travel portal recently switched to a high inward traffic mode and after that there production servers were crashing every 4 hours during peak time. So, diagnosed and resolved there critical issues along with performance tuning of there product.

Oct 2007 – Oct 2008 · 1 yr

• After the acquisition by Oracle, took the product to higher levels by creating more business value of the product, spreading it to newer platforms, which helped the product to generate immense business as brand Oracle.
• Presented the tool "AD4J" in an all industry conference "Developer Summit 2008" which got lot of traction for the product from various people of the industry.

Jul 2006 – Sep 2007 · 1 yr 2 mos

• I was the initial core member of the startup and was there with it from its creation to successful completion. Being the lead developer and architect for substantial part of the product, I developed a performance monitoring tool for java based products which stood out from all the existing tools in the market and generated high business value for the users.

Jun 2004 – Jun 2006 · 2 yrs