Akshay Katageri

About

Built and shipped a production LangGraph-based multi-agent LLM analytics/decision copilot that answers questions across SQL/BI systems and unstructured docs, emphasizing grounded, tool-verified outputs with citations and confidence gating. Deep hands-on experience with orchestration (LangGraph, CrewAI, OpenAI Assistants, MCP) plus real-world latency/cost optimization (vLLM batching/KV caching, speculative decoding, quantization) and rigorous eval/observability. Partnered closely with business/ops stakeholders to deliver explainable reporting automation, cutting manual reporting time by 50%+.

Experience

Education

Key Strengths

• Built and deployed a production multi-agent LLM analytics/decision copilot integrating SQL/BI and unstructured documents
• Strong grounding and trust approach: hybrid retrieval + tool-verified outputs + self-correction loops
• Performance optimization for LLM systems (batching, vLLM continuous batching/KV caching, speculative decoding, quantization, token reduction)
• Reliability at scale via structured agent handoffs, retries, and fallback paths (vs. brittle prompt chains)
• Evaluation-driven engineering: gold datasets, adversarial testing, simulation/load testing, continuous eval with human review
• Production observability and measurable performance management (tracing/logging/metrics; monitoring drift and cost regressions)
• Effective cross-functional delivery with non-technical stakeholders; co-designed guardrails and transparency features
• Demonstrated business impact: reduced manual leadership reporting time by 50%+ and increased adoption through explainability