Ansh Krishna

About

Built an autonomous decision analytics LLM agent for end-to-end tabular binary classification, using RAG (FAISS) to retain context across multi-step queries. Deployed as a FastAPI service with production-style reliability features (schema-aware validation, fallbacks, retries, structured outputs) plus offline/online evaluation and monitoring to reduce analysis time and improve consistency versus stateless approaches.

Key Strengths

• Shipped LLM-powered analytics agent to production as a FastAPI service
• Architected end-to-end multi-step agent workflow for tabular ML (ingestion → EDA → preprocessing → feature engineering → modeling → evaluation → insights)
• Uses LLM as coordinator with deterministic Python functions for predictable execution
• Designs robust guardrails: strict schemas, validation checks, fallbacks, and retries at each stage
• Built offline eval suites and online step-level monitoring with stage-based error analysis
• Improved reliability by moving from free-form intermediate outputs to strict structured schemas with validation gates
• Handles messy/unknown schemas via profiling, type inference, null-rate/cardinality checks, and schema-aware transformation validation
• Implements idempotent pipeline stages with structured intermediate outputs keyed by dataset and step state

Experience

Education

Skills

Languages