Dec 2023 – Sep 2025 · 1 yr 9 mos · Bangalore Urban, Karnataka, India · Hybrid

• 🔶 LLM & Computer Vision NeuralNet Pipeline Optimization for Mobile SoCs
• Enabled Computer Vision and NLP NeuralNets for Samsung chipsets (Exynos), using chipset specific
• optimizations, while maintaining accuracy and latency constraints
• Independently refactored the legacy codebase for CUDA and GPU compatibility in 1 week, a task initially estimated to take 2–3 months with a full team
• Reduced model support lifecycle by streamlining test workflows, resolving integration bottlenecks, and implementing development process optimizations
• 🔶 Secure LLM Chatbot for Technical Documentation
• Architected a fully on-premise RAG system for internal documentation retrieval (Confluence)
• Enabled role-based access control to ensure document-level security
• Improved latency via LLM Quantization and smart document chunking
• Integrated with internal workflow system for seamless information retrieval
• 🔶 LLM Copilot for Codebase Analysis and Vulnerability Remediation
• Built a code-generation and ramp-up assistant using with custom system prompts and
• in-context learning
• Enabled code analysis support via embedding-based chunking of internal Git repos
• Embedded integration with internal mirrors of NVD’s CVE database to cross-reference OSS dependencies.
• 🔶 AI-powered Solution to Automatically Resolve Code Merge conflicts
• Used QWEN Coder to develop a custom merge driver for git, improving team productivity and minimizing integration bottlenecks
• The solution was adopted across the team

Mar 2023 – Nov 2023 · 8 mos · India · Hybrid

• 🔶 Led technology development, R&D, and client engagement across bioinformatics initiatives
• Designed and deployed three bioinformatics pipelines for processing terabyte-scale biological assay data on client-facing servers
• Integrated LLMs and Deep Learning models into production pipelines for enhanced interpretability and automation
• Managed team of five, overseeing daily tasks, reviews, and client-facing deployments
• 🔶 LLM-Powered Chatbot for Multi-Omics BioTech Datasets
• Led the team to develop a natural language chatbot using LLMs to interact with large-scale BioTech
• datasets
• Integrated chatGPT API to generate SQL queries to answer user prompts
• Enabled cross-modal querying across RNA-seq, ATAC-seq, and clinical metadata to support exploratory research

Sep 2022 – Feb 2023 · 5 mos · India

• 🔶 Image to Latex Optical Character Recognition:
• Built a novel transformers based OCR system for mathematical equations in PyTorch
• Achieved 87% BLEU score on image data with noisy background
• Researched and designed improvements to reduce inference time of the NeuralNet by a factor of 2-10 using linearized attention
• 🔶 Traffic Surveillance for Railway Crossing
• Built a vehicle detection and tracking system using ResNet-50 (QAT + mixed precision) and LoRA-adapted DETR for CCTV traffic analysis
• Reduced model size by 4× while improving latency by 6×, still achieved 82% MOTA for vehicle tracking

Feb 2020 – Jun 2022 · 2 yrs 4 mos · United States

• 🔶 Attention Mechanism for Biological Assays:
• Lead the effort to build a nextflow pipeline to create the first dataset of matching mouse and Human DNA features (enhancers) using Deep Learning.
• Achieved 83% accuracy in predicting enhancer-promoter links. The model used LSTM attention and 1-D convolution to predict expensive ground truth assay outputs from inexpensive assays like ATAC-seq
• 🔶 Deep Learning Model for COVID:
• Designed the architecture for using 2D CNN model on patient CT scan images
• Reduced the data requirement for CT scans by a factor of 5, by using transfer learning from 2D CNNs to 3D models
• Reduced false positives by 20% using the 3D architecture

Jul 2018 – Dec 2019 · 1 yr 5 mos · United States

• 🔶 Dimensionality Reduction for Gene Expression Data:
• Invented an online matrix factorization algorithm, with proven stochastic convergence guarantees, for computing basis elements representative of the underlying gene expression signature
• Published in NeurIPS 2019, the algorithm reduced computation requirements by a factor of 100 while
• speeding up convergence

Jun 2017 – Aug 2017 · 2 mos · Hong Kong

• SOLVED a 75 year old open problem in the area of information theory and group testing.

Jun 2016 – Jun 2018 · 2 yrs

• Worked on research problem in distributed data storage, security and estimation using tools in information and coding theory, probability, and combinatorics.

Jun 2014 – Sep 2014 · 3 mos · San Francisco Bay Area

• Proposed performance improvements for current and next generation 802.11n/ac WiFi chips leading to 10% improvement in WiFi coverage. The improvements were implemented and tested in python in-house.

Jan 2012 – May 2013 · 1 yr 4 mos · Kanpur Area, India

• ◦ Proposed fast, provably optimal schemes for Markov Decision Processes in wireless networks
• ◦ Proposed a provably optimal Expectation Maximization (EM) based algorithm to determine
• the sleep schedule for power limited devices
• ◦ Characterized performance of distributed sensing in networks based on stationary Poisson
• Point Processes

Sep 2011 – Dec 2011 · 3 mos · Bangalore, India

• ◦ Designed cyclic codes based memory controllers on Analog micro-processors leading to a 27% improved robustness from errors.

Jun 2010 – Jun 2010 · 0 mo · Kanpur Area, India

• Designed and conducted a week long workshop for Simulink, and MATLAB for 2nd year
• undergraduate students