Laksh Krishna Sharma|

• Led LLM cost optimization, reducing token consumption by 35–45% through prompt compression, JSON to TOON migration and context summarization.
• Migrated AI workloads from Google Gemini to AWS Bedrock, cutting inference costs by ~30%.
• Designed context-engineering pipelines to summarize conversations with 50%+ lower context.
• Architected in-house Astro Engine API, replacing paid providers, reducing API costs by 85%+.
• Built GraphQL APIs using FastAPI and pyswisseph for Vedic astrology at production scale.
• Designed AWS infrastructure (VPC, subnets, NAT Gateway, ALB) for Report and Astro Engine APIs.
• Containerized with Docker, published to ECR, deployed via ECS Fargate for horizontal scaling.
• Built revenue-generating report platform using Next.js and Playwright for paid sales.
• Optimized report prompts, lowering per-report cost by ~25%.
• Scaled async processing with BullMQ and Redis for high-throughput execution.
• Deployed on AWS EC2 with Nginx, PM2, Bastion Host, ALB for secure access.

• Designed agentic RAG AI assistant for SuperSalesIQ CRM, enabling queries over 10k+ leads.
• Built pipelines using LangChain, LangGraph, FastAPI, Milvus, MongoDB, Docker, achieving ~30% faster responses.
• Engineered vector-search workflows, improving accuracy by 25%+.
• Developed backend for SuperConnect, syncing 1k+ LinkedIn connections/session to Sheets.
• Designed authentication, extraction, webhook layers for real-time CRM enrichment.

• Modernized APIs using async/await, improving throughput by 40%.
• Migrated Flask to FastAPI, reducing latency by ~25%.
• Integrated LLM endpoints with LangChain, reducing manual processing by 20%+.
• Refactored backend, cutting development turnaround by ~30%.

Deprecated wait policy in favor of wait for completion, aligning with Amazon operator standards.

Fixed EmrCreateJobFlowOperator deferral logic to respect wait policies, ensuring submit-only jobs no longer defer unintentionally.

Contributed to the apache/iceberg-python repository by enhancing the test_version_format() function, fixing error messages, and adding additional tests to provide clearer guidance for version mismatches.

Contributed to the MDAnalysis/mdanalysis repository by enhancing the AtomGroup.unwrap() method, fixing error messages to provide clearer guidance for missing bond definitions.