Nov 2007 – Feb 2013 · 5 yrs 3 mos · Bengaluru Area, India

• Principal Investigator & Algorithm Design Lead | Strand Life Sciences
• Lead the design and development of intelligent data processing algorithms, specializing in extracting structured insights from massive, unstructured datasets (publications, forums, and internal memos).
• Key Focus Areas:
• 1. Biomedical NLP & Information Extraction (2012–2013)
• Architected a high-scale pipeline to mine decades of PubMed publications for gene-disease associations and genetic variants.
• Developed custom entity and relation extraction tools to transform millions of abstracts into structured, searchable databases for researchers.
• 2. Competitive Intelligence & NER Innovation (2009–2011)
• Dictionary Matching: Engineered an advanced matching system handling symbolic, inflectional, and morphological variants to maximize recall.
• Word Sense Disambiguation (WSD): Developed a supervised WSD technique that generates its own training sets by leveraging neighboring dictionary terms, solving the "small training set" problem.
• Entity Extraction: Designed a "Double Bootstrapping" algorithm to autonomously learn new terms and patterns from unannotated data.
• Tech Stack: Managed full-stack deployment (Java/Grails), indexing (Solr/Lucene/Zoie), and custom visualizations (ExtJS/JavaScript).
• 3. Sentiment Analysis & Opinion Mining (2007–2008)
• Developed algorithms to associate opinions with specific product features by modifying NLP grammars for noisy data (blogs/forums).
• Built contextual entity resolution algorithms to attribute sentiments to specific product models, even when not explicitly named in a sentence.
• Additional Projects:
• Historical NLP: Extracted non-standard location names from colonial-era documents.
• Knowledge Discovery: Built discovery tools for Indian medicinal herbs and word-association measures for intent-free data browsing.

Jun 2004 – Jul 2007 · 3 yrs 1 mo · Bengaluru, Karnataka, India · On-site

• Early work on adaptive, unsupervised video understanding for large-scale surveillance systems.
• Led development of an end-to-end moving object detection, tracking, and classification system, with components deployed in Honeywell products.
• Led design of people and pedestrian detection algorithms for surveillance scenes with high occlusion, non-rigid motion, and wide appearance variation.
• Developed a part-based detection framework using AdaBoost to learn discriminative body parts, enabling fast and robust detection under partial occlusion.
• Designed motion detection pipelines using Gaussian Mixture Models, augmented with optical flow constraints to suppress false positives from background motion (e.g., foliage, water).
• Applied Markov Random Fields (MRFs) to enforce spatial consistency in object detection across frames.
• Architected and implemented particle filter–based tracking for robust prediction and continuity under noise, occlusion, and appearance changes.
• Led exploration and evaluation of adaptive appearance models balancing rapid object variation with robustness to environmental noise.
• Co-developed an RBF-based classifier to categorize tracked objects as human, vehicle, or non-target.
• Drove the full system lifecycle: data collection and ground truthing, algorithm selection, metric definition, and performance evaluation.
• Served as principal investigator and primary contributor to the tracking architecture, guiding work from research through product integration.

Apr 2003 – Jun 2004 · 1 yr 2 mos · Bengaluru, Karnataka, India

• Led an early content-based retrieval project for paper manufacturing defects, helping manufacturing teams quickly identify recurring defect patterns and reduce manual inspection effort