Hands-on software engineer at Google specializing in NLP, Machine Learning, and Large-Scale Information Retrieval. Built chat and voice agents for 100k+ lead-gen advertisers using Gemini, Cloud, and GTP. Designed retrieval infrastructure for serving answer embedding models with 300M+ document indexes and 200M DAU. Skilled in C++, Python, Agents Harness + Evals, LLM-as-a-Judge, PyTorch, TensorFlow, MapReduce, BigTable, Query understanding and Web-Search ranking and retrieval.
[Ongoing] Building chat and voice agent for 100K+ 3P advertisers to improve lead nurturing and qualification with custom tools using Gemini, Cloud ADK, Vertex AI (Search grounding), and GTP.
Built Search Home page recommendation for Play Store - generated high-quality recommendable queries using Google PaLM 340B model with safety evaluation.
Redesigned Play Store suggest (autocomplete) indexing and serving: 200M+ query indices, 20ms p99 latency, personalization model with user signals at 80K QPS (6B+ queries/day).
Founding engineer of Android System Intelligence team. Led design and implementation of Universal Search - a new search surface in Pixel Launcher enabling on-device search across installed apps, contacts, settings, screenshots, and files. Designed ranking and indexing architecture balancing low latency, relevance, and privacy. Led the next app-predictions feature using DNN inference on-device. These launches rolled out to 8M+ pixel devices.
Built Deep Neural retrieval system with 300M+ question embeddings, serving 10% of web-search with low latency; powers “People also ask” feature enhancing user engagement.
Built Fact checking feature in Google Search including modifying indexing and ranking - fact-checking news articles to combat misinformation, resulting in 50k+ fact-checked articles with 4+ billion annual impressions.
Launched several query understanding improvements (synonyms, salient-terms) in web-search ranking with significant IS@5 (Information-Satisfaction) gains.
J Sundararaj, A Vyas, B Gonzalez-Maldonado, “Automated LaTeX Code Generation from Handwritten Math Expressions Using Vision Transformer”, arXiv preprint, 2024
J Kumar, K Jayakumar, J Sundararaj, “ESMCrystal: Enhancing Protein Crystallization Prediction Through Protein Embeddings”, Proceedings of the Conference on Computational Intelligence Methods for Bioinformatics and Biostatistics, 2024
J Sundararaj, J Jayanth, P Bhattacharyya, “Opinion Summarization using Submodular Functions: Subjectivity vs Relevance trade-off”, 16th International Conference on Intelligent Text Processing and Computational Linguistics, 2015. Won Best Verifiability, Reproducibility, and Working Description award
J Jayanth, J Sundararaj, P Bhattacharyya, “Monotone submodularity in opinion summaries”, Proceedings of the Conference on Empirical Methods in Natural Language Processing, 2015
Your Android Phone May Soon Predict Your Habits to Suggest Helpful Apps
Pixel Launcher search showing results from Google Photos, Files, Contacts, and Settings
Finding fresh, helpful information through featured snippets