Open to opportunities

Anand Roy

@anandroy

Research Engineer building production GenAI—RAG, multi-turn dialogue, and LLM orchestration to cut costs and ship tools users rely on.

India

Message

What I'm looking for

I’m looking for a team where I can own production GenAI systems—shipping RAG and multi-turn dialogue that measurably improves quality and reduces cost. I value rigorous evaluation, observability, and fast iteration from prototype to rollout.

I’m a Software Engineer focused on building production Gen AI systems, especially multi-turn dialog systems, RAG pipelines, and LLM orchestration. At Sandvine, I shipped tools used by 300+ users and reduced token costs by 70%, while keeping quality measurable through evaluation and benchmarking.

In my roles, I built an LLM-powered Case Trainer app that evaluates and scores user responses and was adopted by 5 teams within a month. I migrated retrieval from a flat vector store to a graph-based architecture for knowledge-grounded conversations (cutting query latency by 30%), and I built an end-to-end RAG pipeline ingesting 25,000+ chunks for natural language querying of technical docs. I also built an automated NLP pipeline over 17,000+ support tickets with LLM-based classification and added observability dashboards in Grafana for the Customer Success org, backed by CI/CD for LLM microservices using GitLab.

Experience

Work history, roles, and key accomplishments

Current

Software Engineer-2

Current

Sandvine

Jan 2026 - Present (5 months)

Built an LLM-powered Case Trainer multi-turn conversational agent that evaluates and scores user responses, adopted by 5 teams within a month. Migrated retrieval from a flat vector store to a graph-based retrieval architecture, cutting query latency by 30% and improving relationship mapping across technical docs.

RAG Graph Retrieval Vector Databases Prompt Engineering Semantic Search

Software Engineer

Sandvine

Jul 2024 - Jan 2026 (1 year 6 months)

Built an end-to-end RAG pipeline ingesting 25,000+ chunks with embedding models and hybrid semantic search, enabling natural-language querying used by 300+ users. Developed an AI toolkit over Salesforce support data that generates SQL, calls external tools via function calling, and produces contextual summaries; reduced token consumption by 70% through prompt optimization and model evaluation.

Embeddings Semantic Search Function Calling SQL Generation Prompt Engineering LLM Evaluation

Software Engineer Intern

Sandvine

Jan 2024 - Jul 2024 (6 months)

Trained ML models to score network application performance on a 0–5 scale using decision trees, random forests, and XGBoost on structured network data.

Machine Learning Decision Trees Random Forest XGBoost Model Training Feature Engineering Supervised Learning