Venkat Sai Konduru

Continued pretraining in Small Language Models (SLMs) using RS-LORA and evaluated its impact on domain-specific adaptation. Demonstrated that SLMs like LLaMA and Gemma adapt well to proprietary datasets but extended pretraining leads to notable knowledge retention issues. Published research demonstrating knowledge retention challenges in SLMs post continued pretraining, accepted at ACDSA 2025.

Worked on building a virtual assistant for Credit Agricole. Developed pre-meeting and post-meeting synthesis use cases to extract actionable insights from customer-agent interactions.

Researched on evaluating the knowledge loss using continuing pre-training on SLMs. Utilized raw data to derive actionable insights, directly impacting business outcomes. Leveraged expertise in CPU inference to optimize performance and enable efficient local deployment of LLMs.

Fine-tuned transformer based encoder models to extract domain specific representation. Developed a combined model using CNN and GRU to enhance prediction and explanation. Designed an unsupervised summary algorithm using a knapsack based approach from the summaries of the role labelled sentences in the document.

Developed a distributed logging server for logging and log-retrieval of various topics. Architected the distributed server consisting of HTTP servers, main server, replica server and multiple brokers. Implemented Raft Consensus to support log replication.

Grade: 9.14 CGPA

Pursued a Master of Technology in Computer Science and Engineering, achieving a CGPA of 9.14. Focused on advanced topics within the field.

Grade: 8.45 CGPA

Completed a Bachelor of Technology in Computer Science and Engineering, earning a CGPA of 8.45. Gained foundational knowledge and practical skills in computer science.