Arun Brahma

Walmart Global Tech

• Developed a Shopping Assistant using multi-agent architecture and MCP protocol, orchestrating Text-to-SQL, Graph and RAG sub-agents for search, analytics, and automation.
• Increased conversion rate by 37% across pilot categories through reliable intent routing, context retrieval, and automated workflows.

Citi Bank

• Built an enterprise-grade Text to SQL conversational analytics platform using fine-tuned Llama model that would generate insights on complex financial data.
• Automated SQL query generation to boost analyst productivity by 65% and ensure compliance with financial data governance.

Carelon Global Solutions

• Designed a scalable hybrid RAG application on health insurance policy documents using Qdrant for semantic retrieval and BM25 for sparse retrieval.
• Achieved 23% improvement in customer satisfaction feedback compared to traditional chat support.

Accenture

• Implemented a scalable two-tower embedding model architecture to generate candidate product recommendations using user features and product features.
• Improved click-through rate by 47% using new product recommendation pipeline thereby positively impacting the organization’s top line.

• Developed an open-source Python library translating scanned standard PDFs into markdown via Vision LLMs, capturing text, tables, images, LaTeX.
• Benchmarked Vision Parse against Microsoft’s MarkItDown across diverse PDFs, delivering 0.88 markdown accuracy, outperforming MarkItDown’s 0.52 by 69%.

• Trained a Falcon-7B LLM using Low-Rank Adaptation of Quantized LLMs (QLoRA) on mental health dialogues, stripping PII.
• Achieved a ROUGE-1 score of 0.37 in generating relevant and empathetic responses for unseen mental health queries.

Wrote a book for young Indian professionals on guarding the invisible salary of attention, energy, judgement, time, and calm that quietly decides what the visible paycheck becomes in the AI age.

Built a no-login browser chess app to play a tuned engine across ten levels or practice with live Stockfish analysis, with a Gemini Vision feature that turns a screenshot of any board into an editable position.

Built a free, open-source macOS dictation app that turns speech into text entirely on-device on the Apple Neural Engine, with no cloud, accounts, or telemetry.

Take part as a recurring Google UX Research participant since 2024, sharing feedback on early AI/ML and Google Cloud products including BigQuery, Notebooks, and the Gemini CLI.