Days -> Min
Multi-Agent NL Analytics
LangGraph and Bedrock answer plain-English questions with Athena SQL over S3.
Companion RAG cut research time about 50%. A direct partnership with leadership expanded the team.
Hi, I'm
Sureshkumar Mahalingam
Lead AI Engineer & AWS Solutions Architect
I ship production AI agents and RAG systems on AWS, and the data platforms they run on.
Professional
Days -> Min
LangGraph and Bedrock answer plain-English questions with Athena SQL over S3.
Companion RAG cut research time about 50%. A direct partnership with leadership expanded the team.
77K requests
Two agents on AgentCore and Strands: an NL-to-SQL API other teams call, and RAG over 77K service requests.
Drafts work orders. A verified-answer loop raises accuracy without retraining.
2.2h -> 36m
Nearly 600 tables, cloud to on-prem SQL Server, on Step Functions and ECS Fargate.
Row-count verification. Recurring failures brought to zero.
3h -> 15m
Config-driven PySpark and DuckDB across S3, PostgreSQL, Redshift, and Iceberg.
Estimated 30-50% lower run cost. A Next.js dashboard cut troubleshooting 60-80%.
539 / day
Legacy SQL Server landed daily as S3 Parquet, Athena, and Glue.
Automated drift reports replaced manual reconciliation.
70% less compute
Six Spring Boot services and Next.js on ECS Fargate and Aurora PostgreSQL.
Engagement up 30%. Load times down 20%. Scheduled shutdown cut non-production compute about 70%.
Personal
Gift-store assistant on Bedrock Converse with a custom Lambda tool loop.
Tool loop for search, policy Q&A, and cart - KB + S3 Vectors for citations
Same assistant on Bedrock Agents - managed orchestration and action groups.
OpenAPI action-group Lambda with Knowledge Base policy retrieval
Same assistant on AgentCore Runtime, Gateway MCP tools, and Memory.
Runtime Converse loop with Gateway tools and short-term session memory
Same assistant on Bedrock Mantle Responses API with server-managed state.
OpenAI-compatible Responses API with stateful previous_response_id turns
PDF upload to Textract, Bedrock structuring, and hosted portfolio output.
Event-driven S3 pipeline - Bedrock JSON model to Jinja portfolio on CloudFront
66x / 12x
API Gateway + Lambda + DynamoDB, tuned across six memory configs.
512 MB vs 128 MB - faster and cheaper; CDK runtime + Step Functions tuner
Fitbit-style CDK lab: Cognito, IoT, SQS, analytics, and live insights.
Live dashboard for steps, activity, and heart rate with WAF-hardened edge
Compares in-memory, DynamoDB, and Redis when Lambda traffic fails over.
Mid-session failover visualizer - carry-forward vs session loss
Kinesis on-demand -> Lambda -> DynamoDB. Live AWS demo paused to control stream-hour cost.
Partial-batch retries and atomic view / cart / purchase counters
0-100
Ghost-job detector for Greenhouse/Lever URLs with a live paste-a-URL check.
Risk score from age, reposts, and careers cross-check (~$3-10/mo)