Amazon

- Supervised the development of sophisticated music recommender systems addressing various stages, including cold start, warm start, and dynamic "next best action" suggestions.
- Took the lead on the proof of concept for an item-to-item music recommendation model, in a strategic collaboration with Amazon.com's retail platform.
- Spearheaded rigorous A/B testing campaigns to empirically assess and optimize the recommender's enhancements, ensuring data-driven decisions and superior user experiences.

Machine Learning Leader

Indian Institute of Technology, Patna

Pursuing fundamental research in NLP, Vision, and Multimodal AI.

Research Fellow

University of South Carolina

Pursuing fundamental research here in NLP and Multi-modality. Publish in leading conferences such as EMNLP

Oracle

- Using Machine Learning to model server capacity data to better be able to predict outages
- Lead the charter of building out a recommendation system for predicting object storage space based on customer usage patterns.
- Lead efforts to predict server health by applying LogBERT to understand anomalies.
- Maintained a set of curated ML workloads comprising of training, fine-tuning and inference jobs to benchmark server performance levels.
- Oversaw the development of outlier detection for success of our current latency and storage models.

Cloud AI Engineer Leader

Palo Alto Networks

- Managed a team to build out a new Distributed, Microservices architecture for licensing  Palo Alto Networks cloud cybersecurity products.
- Directed and led a dynamic team to craft and implement a state-of-the-art Distributed Microservices architecture, vital for efficiently licensing Palo Alto Networks' cloud cybersecurity solutions. This framework places an emphasis on trustable, ethical AI practices to ensure transparent and unbiased decisions.
- Pioneered the incorporation of AI technologies into our cybersecurity offerings, ensuring not only advanced threat detection but also a commitment to ethical and transparent AI methodologies that users can trust.
-  Orchestrated the robust implementation of Role-Based Access Control (RBAC) for apps.paloaltonetworks.com. Leveraged AI-driven insights to bolster platform security, all while ensuring AI models remained transparent and free from biases.

Engineering Leader

Apple

- Entrusted with the responsibility of shaping the future of Apple's intricate internal microservices.
- Expertly harness Java 8, SpringBoot, and Kafka to design and develop robust, scalable solutions.
- Utilize MongoDB to guarantee data consistency, reliability, and swift retrieval across our microservices landscape.
- Consistently ensure our software solutions align with Apple's unparalleled benchmarks of innovation, efficiency, and user experience.

Senior Software Engineer

Razorfish

Developer worked on launching several ecommerce sites for a variety of clients

Software Engineer

University of Maryland College Park

Edited websites to incorporate web accessibility, updated campus websites, created pages for departments on SharePoint, and keept track of web requests.

Technical Assistant

Seasoned Machine Learning Leader specializing in Natural Language Processing and Recommender Systems with industrial experience in large scale companies and a background  in Artificial Intelligence from Stanford University.

🎳 RAG , Finetuning, and Parameter Efficient Finetuning (PEFT)

When adapting large language models (LLMs) to specific tasks, choosing the right technique is crucial. Whether it's RAG, Finetuning, or PEFT, each method has its unique advantages. Here's a detailed look at each:

🎾 RAG https://vinija.ai/nlp/RAG/
◦ Explore everything about RAG from a comprehensive overview to metrics and implementation. Stay updated with the latest in the field, including Retrieval-Augmented Fine-Tuning (RAFT) by Tianjun Zhang et al., and the emerging Multimodal RAG, reflecting the trend towards multimodality, similar to Google's Gemini.

🏈 Finetuning https://lnkd.in/gGfM6k8K
◦ Dive deep into the nuances of finetuning LLMs, featuring insights into recent innovations such as Reasoning with REinforced Fine-Tuning by ByteDance.

⚽️ PEFT https://lnkd.in/gNKxjHz7
• Discover efficient finetuning methods like freezing lower layers and adapting top ones for specific tasks. The article also covers advanced techniques such as LoRA, ReFT, and Low-rank Linear Subspace Representation Finetuning (LoReFT).


⭐️ RAG vs Fine-tuning 
- Choosing between RAG and Finetuning? Consider insights from this study: https://lnkd.in/gcHq6UfE

🔹 Fine-tuning boosts performance across all entities, particularly the most and least popular ones.

🔹 RAG typically outshines fine-tuning when used alone and offers significant benefits when combined with FT.

🔹 The success of both strategies grows with advances in retrieval technology and the quality of synthetic training data.

🔹 Smaller models especially benefit from the combination of RAG and FT, though larger models see less improvement due to their vast pre-trained knowledge.

🚀 Interested in learning more? Consider these resources:

📚 Workshop on hands-on experience: Is RAG Dead: https://lnkd.in/gh3Y3CS2

📚 Data Analyst Certificate by Meta and Coursera : https://lnkd.in/gGc7SVf8

This is written in collaboration with Aman Chadha, follow on Twitter (https://lnkd.in/geUBXFwq) to stay up to date with the lates in NLP and Recommender Systems :)

Taplio

Vinija Jain's Linkedin Analytics

Vinija Jain

What is Vinija talking about?

Vinija Jain's Best Posts (last 30 days)

Want to drive more opportunities from LinkedIn?