Assessed AWS Bedrock Agents; concluded it was not production ready. Architected and built a fully compliant A2A agentic framework. 100% AWS serverless. Dynamic loading of agent cards and tool manifests.
Outcome: Cross-team Agentic AI development framework delivered ahead of time.
Built an end-to-end automated pipeline for processing 100,000 invoices per month, integrating multiple large language models and OCR tools. The project included high accuracy data extraction and a 99.99% up-time.
Outcome: $2m cost reduced to $50k; process lead time improved from 2 weeks to 2 days; error rates dropped by 90%.
Led the build of an OpenAI API based chatbot with a strong guardrails requirement. Secure hand-off to human-in-the-loop support workers
Outcome: Successful handover of platform to clients team in Singapore
I love to evaluate and use the multifarious frameworks and tools
available for building agentic AI solutions.
However, my preferred approach nowadays is to build agentic AI systems
that are a little more
autonomous and
self orchestrated.
For this, I tend to use the lower level LLM APIs directly within my own
architecture wrapped with standard protocols (A2A and MCP) and hosted
within IAC built serverless, scalable, cloud infrastructure.