Jaya Mathan R

• Developed LLM orchestration service for classifying and routing support tickets, generating response drafts using OpenAI tools.
• Established guardrails including prompt templates, policy checks, PII masking, and content filters for regulated responses.
• Implemented feedback loop incorporating agent edits into training dataset with regression tests for continuous prompt evaluation.
• Integrated event-driven processing workflow utilizing SQS/SNS and dashboards monitoring auto-resolution percentage and CSAT impact.

Utilized AWS technologies: Python, OpenAI API, SQS/SNS, Lambda, Step Functions, DynamoDB, S3, CloudWatch, IAM, KMS.

• Developed observability layer for LLM applications, logging prompts, context, outputs, and metrics.
• Implemented offline evaluation suite with golden Q/A sets and automated comparison methods.
• Established cost controls through model routing, prompt compression, and token budgeting per tenant.
• Delivered dashboards and alerts for latency, failure rates, and cost anomalies.
• Utilized AWS technologies including Python, OpenAI API, CloudWatch, and QuickSight.

• Developed streaming pipeline scoring events with Python ML models, triggering alerts in near real-time.
• Introduced OpenAI-based explainability, providing human-readable reasons and feature contributions summaries.
• Implemented risk scoring, deduplication, and case creation workflow, enhancing analyst efficiency with structured summaries.
• Monitored drift and model performance, automating retraining based on drift thresholds.

Utilized AWS technologies including Kinesis/MSK, Lambda, SageMaker, DynamoDB, S3, OpenAI API, CloudWatch, and SNS.

• Developed RAG system in Python utilizing OpenAI embeddings and chat models for PDF/Doc queries with grounded citations.
• Implemented ingestion pipeline, including chunking, metadata extraction, deduplication, and versioning for efficient processing.
• Stored vectors in Amazon OpenSearch (k-NN) and pgvector on RDS/Aurora for optimized retrieval.
• Integrated RBAC and tenant isolation using IAM/JWT claims to ensure data security across teams.

Reduced latency through caching with ElastiCache and asynchronous retrieval methods.

• Established automated evaluation metrics (precision@k, citation coverage, hallucination rate) for system performance assessment.
• Utilized AWS technologies including S3, Lambda/Fargate, API Gateway, DynamoDB, CloudWatch, KMS, and Secrets Manager for comprehensive solutions.
• Employed Terraform/CDK for infrastructure management and deployment efficiency.

• Developed multi-agent analytics copilot in Python utilizing OpenAI function calling for SQL translation and analysis steps.
• Implemented execution layer with SQL allowlist/denylist, row/column-level security, and sandboxed Python execution.
• Orchestrated AWS Step Functions to automate processes from intent detection to report generation.
• Enhanced reliability with features like retries, idempotency keys, query caching, and observability metrics.
• Delivered automated outputs including KPI summaries, anomaly reports, and formatted downloadable files.

Utilized AWS technologies such as Python, OpenAI API, S3, Lambda, and Athena to support analytics functions.

• Developed document compliance checker in Python utilizing OpenAI models for clause extraction and comparison against control library.
• Implemented PII/secret redaction with configurable masking rules based on document metadata.
• Designed two-pass pipeline for structured extraction into JSON schema, followed by rule evaluation and evidence-supported recommendations.
• Incorporated audit logging for tracking execution details and immutable storage for reports using S3 versioning and KMS encryption.
• Created API endpoints and lightweight UI for document upload, review, and export with role-based access controls.

Leveraged AWS services including Lambda, Fargate, API Gateway, and DynamoDB to enhance functionality.