Akshay Kumar C P

Machine Learning Engineer (Senior Consultant)

(Second company)

Summary

• Built Retrieval Augmented Generation (RAG) & Agentic AI systems for client projects.
• Conducted research on Large Language Models (LLMs) and related tools, identifying best practices and optimization strategies for development.
• Led requirement gathering sessions, translating business needs into technical specifications, and formulating effective solution implementation strategies.
• Coordinated with cross-functional teams to ensure alignment on project goals and deliverables.
• Analyzed Request for Proposal (RFP) documents, identifying key technical components for detailed discussions and further development.

Project's

Agentic AI Chatbot

• Design & build Agentic AI workflow having text and visualization modes.
• Incorporated ReAct Agentic pattern from scratch and function calling for building reliable and robust Agentic System.
• Handled team of 5 members under my guidance for solution building.
• Deployed solution in an Azure App Service for accessing Chatbot within organization.
• Technologies: Langraph, FastAPI, Azure services – OpenAI (GPT-4o Model), App service, Key Vault, Storage, MySQL Server DB.

RAG Playground

• Contributed document parsers and AI enrichments for building an enriched vector database for downstream chatbot.
• Technologies: Langchain, Microsoft Azure, Databricks & Generative AI.

RAISE platform – Agentic AI

• Agent based Human-in-loop for monitoring and evaluation.
• Security on Generative AI solutions in terms of Prompt injection attacks.
• Data quality metrics for Gen-AI solutions.
• Technologies: Microsoft Autogen & LLM.

Legal Tender AI Review

• Validate contract documents against compliance rules and perform criticality assessment.
• Utilized RAG based architecture and various prompt techniques for performing Legal tender document reviews.
• Involved in requirement gathering, defining approach, performing multiple experiments under development of the solution.
• 60% high confidence criticality assessment.
• 1.4 M€ / year cost avoidance.
• Technologies: GPT-3.5-Turbo LLM, text-embedding-ada-002 embedding model, Azure Cognitive Search vector store, Langchain & Prompt engineering.

Code Assistant

• Generate Pyspark code for Business documents.
• Utilized open source LLM due to constrain on not utilizing APIs
• Activities performed: Prompt Engineering, Orchestration & Deployment.
• Tech Stack: Databricks GPU Compute, Prompt engineering, HuggingFace LLM models such as mistral, llama 2, etc. , Quantization, Chroma vector DB, Langchain, Ollama etc.

Generate Synthetic Data using Generative AI

• Tabular: Utilized libraries such as CTGAN, SDV, etc for generation.
• Images: Generative models such as GAN, BIG-GAN, stable diffusion, etc.
• Text: Used GPT-2, Mosiacml, Bloom, falcon, etc. for text generation tasks.
• Implemented using Hugging Face, TensorFlow Hub and other frameworks.

Personalized Marketing

• Personalize images using Open-Source Stable Diffusion Models.
• Performed Fine Tuning using Dreambooth Framework.

Natural Language to SQL

• Implemented using Open AI “text-davinci-003” model and Prompt engineering techniques.
• Built UI using Gradio framework for quick demo.

AWS Clean Room

• Take marketing decision without third party cookies.
• Configured collaboration between advertiser and publisher using, S3 and Glue.
• Amperity CDP for Identity resolution (IDR) and customer campaigns
• Technologies: Amperity Platform

Customer Satisfaction Analytics

• Research & analysis on requirements and planning.
• Brainstorming the solution with respect to requirements.
• Generating actionable insights from data and creating reports, presentation and dashboards to take key business decision.
• Perform analysis namely predicting loyal customers using cluster models, CSAT, NPS, SERVQUAL, Recency Frequency Monetary (RFM), Cohort analysis.
• Use python for automation (OOPS, REST APIs, logging, exceptional handling, config file. linting, etc.).
• Performed Sentiment classification and Topic Modelling as part of Satisfaction Metrics
• Technologies: Python & Machine Learning.