Lokesh Jain

About

Built and deployed AcademiQ Ai, a production LLM-based teaching assistant using GPT/BERT with RAG (LangChain + Pinecone) to handle large student notes and generate adaptive explanations/quizzes. Demonstrated measurable retrieval-quality gains (18% precision improvement, 22% less irrelevant context) by tuning similarity thresholds and chunking based on user satisfaction signals. Also orchestrated terabyte-scale, real-time demand forecasting pipelines using Airflow and Kubeflow on GCP with strong monitoring, shadow deployment, and feedback-loop practices.

Key Strengths

• Built and deployed an LLM-powered teaching assistant (AcademiQ Ai) to production
• Solved token-limit/context truncation via chunking + embeddings + Pinecone retrieval using LangChain
• Improved retrieval precision by 18% and reduced irrelevant context by 22% through similarity-threshold and chunk-size tuning tied to user satisfaction
• Designed feedback loop using logged outputs (latency/confidence/user ratings) and weekly aggregation to detect drift and low-confidence patterns
• Experience orchestrating large-scale ML pipelines with Airflow and Kubeflow on GCP (terabytes of transactional data, real-time forecasting)
• Defines success metrics upfront and uses shadow deployments plus drift/latency monitoring for reliable AI releases
• Bridged technical and non-technical stakeholders by translating model metrics into operational KPIs (forecast error tolerance, service-level impact)

Experience

Education

Skills

Languages