Mar 2023 – Sep 2023 · 6 mos

Oct 2020 – Jul 2024 · 3 yrs 9 mos · Pune, Maharashtra, India · On-site

• Led a team of 4 engineers to deliver scalable backend systems, owning architecture and release timelines.
• Built a scalable real-time, event-driven loss detection system ingesting multi-logger streams arriving every
• 500ms, improving monitoring efficiency and reducing supervisory time by 50%.
• Re-architected a core ETL pipeline in Python with Luigi framework reducing pipeline duration by 6x.
• Developed and launched an automated email reporting service using Python and Node.js driving a 10%
• increase in weekly engagement with the core analytics platform.
• Owned the end-to-end development of 50+ RESTful APIs in FastAPI, which enabled the entire suite of
• Manufacturing Line Analytics features and directly supported the company’s main product offering.
• Implemented a new monitoring stack using Prometheus & Grafana, reducing Mean Time to Discovery (MTTD)
• of outages from hours to 5 seconds. This allowed the team to proactively resolve 40% issues before impacting the
• customer.
• Decoupled and dockerised 5 microservices, laying the groundwork for a scalable Kubernetes deployment on AWS
• cloud, transitioning from Internet Information Services (IIS) and PM2.
• Onboarded a cloud migration consultancy with our detailed backend architecture, services, and current AWS
• EC2 based deployment to facilitate a successful full AWS transition.
• Managed software deployment across client systems using shell scripts and daemonized processes, ensuring timely
• updates, minimal downtime, and consistent configurations.

May 2020 – Sep 2020 · 4 mos · Bengaluru, Karnataka, India · Remote

• Improved the technical question-answer system for the TechQA dataset, improving support capabilities.
• Domain Adapted BERT (Bidirectional Encoder Representations from Transformers) embeddings on a technical corpus
• using 3 techniques: Masked Language Model, Next Sentence Prediction, and Sequential Question Generation.
• Developed a language model in PyTorch by combining a multilayered BERT-based span selection model with an
• Attention over Attention architecture-based model, and trained the model on a GPU cluster in New York.
• Achieved a state-of-the-art F1 score of 57.7 on the TechQA dev set, surpassing previous benchmarks.

May 2019 – Jul 2019 · 2 mos · New Delhi, Delhi, India · On-site

• Worked on converting Text Stories into full fledged movie scripts using Natural Language Processing. Used Python and SpaCy extensively.
• Concepts used
• 1. Coreference Resolution
• 2. POS Tagging
• 3. Dependency Parsing
• 4. Named Entity Recognition
• 5. Rule based appraoches