Charith Kandula

About

Built a production LLM/RAG document extraction and game/quiz content workflow using LLaMA 2, LangChain/LangGraph, and FAISS, achieving ~94% accuracy and reducing turnaround from hours to minutes. Demonstrates strong applied MLOps/orchestration (CI/CD, MLflow, Databricks/PySpark), robust handling of noisy/variable document layouts (layout chunking + OCR fallbacks), and practical reliability practices (human-in-the-loop routing, drift monitoring, A/B testing).

Experience

Education

Key Strengths

• Built and deployed production RAG-based document extraction system (LLaMA 2 + LangChain + FAISS)
• Improved processing time from hours to minutes while achieving ~94% accuracy
• Handled 500+ document types via layout-aware chunking and OCR fallback/validation for noisy scans
• Addressed imbalanced data and model drift using fine-tuning, active learning, monitoring, augmentation, and semi-supervised learning
• Designed reliable agent/workflow evaluation with unit tests, synthetic simulations, A/B tests, and metrics (accuracy/F1/latency/error rates) plus drift monitoring
• Implemented LangGraph branching to route low-confidence outputs to validation tools or human review
• Effective collaboration with non-technical stakeholders by translating requirements into measurable metrics (accuracy, turnaround time, auditability)
• Designs scalable backend architectures with clear layering (API, data, security, observability)
• API reliability focus: explicit contracts (OpenAPI), stable schemas, and edge-case handling
• Migration/refactor execution using strangler pattern, feature flags, and incremental rollout
• Strong data integrity practices: constraints, transactions, dual read/write, restartable backfills
• Improves robustness by identifying failure modes (duplicate job creation) and implementing idempotent endpoints
• Production security implementation with JWT/OAuth and multi-tenant access patterns