Scaling and optimising LLM inferencing pipelines for Open Source Models on GPU clusters.

Building SuperCoder 2.0, an Open Source autonomous Software Development System

2) Research - Part of the core research team at SuperAGI. - working on model Finetuning techniques - Instruct FT, DPO, etc. - Dataset creation for chat and instruct models. - Agentic Frameworks facilitating multi agent collaborative chat applications. 1) Development of SAM-v1, small agentic model based on Mistral 7B. - Development of Explanation traces to fine tune mistral 7b - Used Yi-34B-Chat, Falcon - series, and Mixtral to generate Rationales - SAM achieved performance comparable to GPT 3.5 and outperformed Orca on GSM 8k. - The aim was to fine tune Mistral 7b with explanation/reasoning traces of better models to achieve better problem solving capability. Blog : https://superagi.com/introducing-sam-small-agentic-model/HuggingFace : https://huggingface.co/SuperAGI/SAM

2) Working in AI/ML to develop POCs and implementing client requirements for automation using Python. Working with frameworks like TensorFlow and PyTorch. Developing Semantic search system for closed domain Question answering using LLMs. Using different open source tools to quickly develop NLP and Document Understanding application. Along with working on OCI AI Services to deliver to client needs. 1) Worked with client to analyze the functional requirements and migrate SOA integrations to OIC Gen2. Analyzed SOA composites for ERP and CRM based systems.

Worked on Deep Learning algorithms and used frameworks like Nvidia Nemo, Nvidia Tao, etc.