Mohit Goyal

Built and deployed production-grade AI systems at scale. Developed agentic AI features (artifacts, agent skills, multi-agent orchestration) and shipped them to production at C3ALabs, focusing on reliability, latency, and real-user usage.

Designed and productionized advanced RAG pipelines using the latest Google File Search APIs, handling real-world issues like document chunking, retrieval quality, and response grounding.

Core contributor to applied AI R&D. Worked in the R&D team, experimenting with and integrating cutting-edge tooling such as Claude skills, MCP-based tool routing (Composio), and multi-agent coordination patterns.

• Co-developed GRIT(Geometric Reprojection Instruction Tuning), a novel instruction tuning framework that fine-tunes
¡1% of LLM parameters (0.997%) while outperforming full fine-tuning and LoRA on benchmarks (BLEU ↑26.2, ROUGE-L
↑34.9 on Alpaca).
• Fine-tuned multiple LLMs including GPT-2 (355M), LLaMA-3B, and Mistral-7B, demonstrating up to 30% parameter
savings without performance loss, reducing compute and memory costs by 40%.
• Conducted rigorous ablation studies comparing GRIT vs. LoRA, QLoRA, and AdaLoRA, validating 2.1 BLEU and 2.5 ROUGE-L gains with GRIT’s geometry-aware updates.