Mohit Dulani

Experience

1. Lead Ai Engineer

   Astrosure Ai

   Aug 2024 — Present

   AI Lead | AstroSure AILed the architecture and deployment of Agastyaa, an AI-powered Vedic astrology oracle integrating LLMs, multi-modal retrieval pipelines, and graph databases to deliver real-time, personalized insights on kundli, panchang, and life decisions like career and relationships via 24/7 chat features.Directed post-training RLHF workflows with Supervised Fine-Tuning (SFT) and GRPO on open-source LLMs using PyTorch Lightning, while fine-tuning models like Gemini 2.0 Flash for routing logic and GPT-4o for answer generation to boost accuracy and user personalization.Designed hybrid retrieval systems combining Qdrant vector search, Neo4j GraphRAG with Cypher queries, and LangChain/LangGraph RAG pipelines, integrated with a Django orchestration layer using Pydantic schemas and a 10-category classification model for dynamic query routing.Developed a custom model load balancer to manage high request volumes and distribute inference across multiple LLM backends (e.g, DeepSeek, Groq, Cerebras), incorporating health monitoring, failover, and an internal A/B testing tool for real-time on-site model comparisons and optimizations.Curated 14K+ domain-specific Q&A datasets via Hugging Face and integrated a MongoDB user state engine for personalized dialogues with async follow-ups and rule-based/semantic retrieval hybrids, achieving <8s latency and reducing token usage from to ~60 per request (90%+ efficiency gain).

Education

• Indian Institute of Technology Jammu

  BTech, Minor in Computer Science with Majors in Materials Science