Randheer Ramesh K

• • Designed an AI root-cause investigation agent for traders and developers to diagnose distributed system failures from timestamps or log lines using Anthropic APIs, agentic SDKs, custom tools, Loki, logcli, and structured investigation workflows.

• • Built a chat frontend for the investigation agent, making complex log and system analysis accessible to engineering and trading users; reached approximately 94% successful investigation quality on supported issue types.

• • Built observability and data pipelines across compute infrastructure using Prometheus, Grafana, Alertmanager, Vector, and Loki for metrics, alerts, and log aggregation.

• • Configured secure Slurm-based compute infrastructure for ML training and data processing, and built custom agentic skills for log analysis using logcli, S3 data access, and automated processing.

• • Redesigned a GenAI WhatsApp banking support assistant for one of India's largest banks, improving response speed by 70% while strengthening security and reliability for customer-facing conversations and direct client communication.

• • Introduced graph-based agent execution with input/output guardrails, hallucination detection, conditional routing, and

retrieval/context optimization using CrewAI, LangChain, OpenAI-compatible APIs, and self-hosted models.

• • Benchmarked and integrated AWS Bedrock Guardrails, NVIDIA NeMo Guardrails, custom safety checks, and self-hosted models for prompt safety, topic safety, content safety, and response grounding.

• • Designed a YAML-driven conversational SOP framework that converts workflow definitions into executable agent workflows with data fields, function nodes, agent nodes, transitions, interrupt/resume, outcomes, and state history.

• • Reduced new SOP agent implementation time to within a day by replacing repeated custom agent coding with declarative workflow configuration, validation, reusable nodes, and traceable execution.

• • Built a FastAPI workflow management API and React/TypeScript AI flow builder to visually create workflows, connect transitions, inspect nodes, generate mock functions/agents with LLMs, replay traces, and visualize state.

• • Built CD4ML, Kubernetes CI/CD, security scanning, and static analysis improvements, reducing build times by 30% and vulnerability detection time by 50%.