Developed a RAG chatbot using Pinecone and GPT4 for a UK-based bank, enhancing customer engagement, promoting eco-friendly banking practices, and responding to 60% of customer queries without manual intervention.Developed a RAG chatbot using Pinecone and fine-tuned Llama-3 enabling user interaction with over 1 million judicial documents. Used FastAPI for integration with the main application.Designed and implemented an AI-driven learning hub using Llama-2 and ChromaDB, and built… Show more Developed a RAG chatbot using Pinecone and GPT4 for a UK-based bank, enhancing customer engagement, promoting eco-friendly banking practices, and responding to 60% of customer queries without manual intervention.Developed a RAG chatbot using Pinecone and fine-tuned Llama-3 enabling user interaction with over 1 million judicial documents. Used FastAPI for integration with the main application.Designed and implemented an AI-driven learning hub using Llama-2 and ChromaDB, and built vector search for multimedia content such as podcasts, YouTube videos, blogs, PDFs, slides, etc.Deployed an AI agent that interacts with customers using audio and video, streamlining room bookings, check-ins, check-outs, and payments. This solution delivers detailed information on hotel amenities and leverages LLMs and talking head models for a seamless guest service experience. This product has greatly enhanced customer satisfaction and operational efficiency, by reducing the check-in process to just one minute. Show less

Led the development and integration of AI-driven products. I contributed from conceptualization through to deployment. Here are the key projects I was involved in:Developed a solution using OCR and fine-tuned Llama-2 model to convert unstructured data from various invoice formats into structured JSON. Reduced costs and manual data entry time by 90%.Developed an enterprise solution to automate the extraction of essential information from Letters of Credit (LC) and the creation of… Show more Led the development and integration of AI-driven products. I contributed from conceptualization through to deployment. Here are the key projects I was involved in:Developed a solution using OCR and fine-tuned Llama-2 model to convert unstructured data from various invoice formats into structured JSON. Reduced costs and manual data entry time by 90%.Developed an enterprise solution to automate the extraction of essential information from Letters of Credit (LC) and the creation of all required compliance documents. Reduced document generation process time from 20 days to 4 days.Recognized for exceptional dedication and contributions to project success, despite working on a limited schedule of 20-50 hours per month. Show less

At Ath3na Technologies, I have played a pivotal role in driving innovation and overseeing the development of our product. Below are the highlights of my contributions:Engineered an RAG-powered chatbot using Vectara and GPT4 enabling users to engage in conversations with their transcripts.Fine-tuned BERT model to classify 26 distinct emotion classes, enriching our platform's emotional intelligence capabilities, and attracting new users.Developed a set of fine-tuned LLMs… Show more At Ath3na Technologies, I have played a pivotal role in driving innovation and overseeing the development of our product. Below are the highlights of my contributions:Engineered an RAG-powered chatbot using Vectara and GPT4 enabling users to engage in conversations with their transcripts.Fine-tuned BERT model to classify 26 distinct emotion classes, enriching our platform's emotional intelligence capabilities, and attracting new users.Developed a set of fine-tuned LLMs to identify various personality traits from transcripts, such as limiting beliefs, victimhood mindset, positivity, etc.Led the architecture, deployment, and ongoing maintenance of our entire data pipeline on AWS. Show less

At Neuris, I was responsible for the initial development of the product. My contributions include:Engineered an automated data collection pipeline capable of extracting Python code from GitHub accounts, including a mechanism for filtering syntactically correct code (using ASTs). Scaled the data collection pipeline to process millions of Python files from public GitHub repos, improving data quality and process efficiency.Fine-tuned a pre-trained GPT-2 model using PyTorch, Hugging… Show more At Neuris, I was responsible for the initial development of the product. My contributions include:Engineered an automated data collection pipeline capable of extracting Python code from GitHub accounts, including a mechanism for filtering syntactically correct code (using ASTs). Scaled the data collection pipeline to process millions of Python files from public GitHub repos, improving data quality and process efficiency.Fine-tuned a pre-trained GPT-2 model using PyTorch, Hugging Face on a dataset that I scraped from public GitHub repositories. Developed custom training methods to efficiently train the model on Google Cloud Platform's TPUs to reduce training costs.Deployed models on GPUs for inferencing. Developed an API using FLASK to serve predictions from the neural network, enhancing the product's accessibility and user interaction, and improving user satisfaction and engagement. Show less

As a Researcher at the Communication Research Lab, I contributed to several government-funded projects, including:Trained deep learning models to navigate UAVs in indoor environments using stereo cameras to predict safe trajectories for real-time obstacle avoidance and led the design and development of an autonomous UAV capable of delivering light packages over a distance of up to 5 km.Trained a CNN to optimize traffic signal control based on real-time traffic density data… Show more As a Researcher at the Communication Research Lab, I contributed to several government-funded projects, including:Trained deep learning models to navigate UAVs in indoor environments using stereo cameras to predict safe trajectories for real-time obstacle avoidance and led the design and development of an autonomous UAV capable of delivering light packages over a distance of up to 5 km.Trained a CNN to optimize traffic signal control based on real-time traffic density data, improving traffic flow and reducing congestion. Show less