Sparsh Rawal

• Executed migration of mission-critical ML/data pipelines from DBX to DAB, enhancing reliability through code refactoring and deployment logic revision.
• Reverse-engineered Kedro-based pipelines, implementing custom DAG-driven orchestration for expedited CI/CD and improved dependency management.
• Automated stale-table cleanup and cluster termination, boosting data governance and quality monitoring across platforms.
• Modernised training, inference, retraining, and evaluation pipelines for Cross-Sell/NBA recommender system using DAB.
• Achieved runtime stability and decreased cloud costs via disciplined DAG orchestration and refined job boundaries.

Engineered and maintained MCP servers (FastMCP) to enable autonomous task execution.

Developed MCP Sandbox, a dockerised environment for isolated Python experimentation.

• Supported deployment of Navari on AWS EC2, integrating Redshift, S3, and Bedrock.
• Identified critical security vulnerability exposing internal clinical data through LLM access.
• Designed Internet Guardrails to enforce security compliance and safeguard confidential data.
• Implemented flag-based heuristics to detect sensitive information effectively.
• Automated internet access restrictions for selected MCP servers to enhance safety.
• Strengthened AI safety posture while ensuring operational flexibility.

• Architected a RAG-based conversational AI system to enhance internal knowledge retrieval.
• Converted extensive, unstructured text repositories into searchable vector embeddings.
• Designed an end-to-end pipeline for optimal semantic document retrieval.
• Chunked domain-specific documents into 800-character overlapping segments for improved accuracy.
• Generated dense embeddings using an OpenAI embedder, storing them in Pinecone vector database.
• Implemented retrieval logic to obtain top-ranked contextual chunks for LLM input.
• Developed prompt templates to enforce compliance and minimise hallucinations.
• Integrated voice-to-text and text-to-speech interfaces for efficient technical troubleshooting.

• Spearheaded GenAI development for forecasting, segmentation, CV, and pricing analytics in UK electrical retail.
• Engineered a multi-agent system with Anthropic Agentic Framework to automate CX analysis and summarise reviews.
• Delivered RAG-powered conversational assistant trained on Trustpilot reviews, enhancing CX issue detection.
• Created Rekognition-based CV model using over 5,000 labelled images, decreasing QA effort by three hours weekly.
• Forecasted washing machine sales via AWS Forecast, achieving 6.3% MAPE accuracy.
• Conducted RFM segmentation on over six million customers using Sagemaker; visualised clusters with Kepler.gl and QuickSight.
• Simulated leader-follower pricing behaviour with Monte Carlo methods for over 4,000 products.
• Authored technical documentation for R&D, facilitating £50K+ tax benefits.

• Delivered targeted ML and NLP solutions through two sequential project engagements.
• Built and deployed ARIMA-based time-series forecasting pipeline for workload demand predictions.
• Cleaned and transformed operational datasets, implementing auto-parameter tuning for optimal performance.
• Provided forecast outputs with evaluation metrics, integrating monitoring scripts for retraining.
• Designed BERT-based semantic matching engine for enhanced document-query relevance.
• Preprocessed internal documents to generate 768-dimensional embeddings for improved accuracy.
• Developed prototype pipelines for clustering text entries and matching queries effectively.
• Created reusable components to support future RAG and search-based systems.

• Contributed to end-to-end feature development for Amdocs Order Management System and CRM product line.
• Developed multiple OMS modules in Core Java, adhering to telecom-grade coding standards.
• Designed and maintained SOAP and REST API integrations for order processing workflows.
• Executed complex Oracle SQL queries for debugging and optimising data access operations.
• Performed enhancements in sprints, writing unit tests and executing regression test cycles.
• Implemented server-side updates via Unix commands and deployment automation scripts.
• Participated in CI/CD processes, managing code integrity with Perforce version control.
• Delivered knowledge-transfer sessions on OMS architecture, APIs, and data models for new team members.