Manav Modi

Played a key role in developing LLM based Chat-bot SaaS agents that can autonomously perform CRUD operations in Ticketing, CRM and Recruiting apps such as Jira, Servicenow, Salesforce, Hubspot, Greenhouse and Workday through natural language prompts thus enabling Ema customers to seamlessly connect and operate on these apps directly through the chatbot.

Led the workflow management in the backend to ensure that the agent can make secure and seamless REST API calls to the SaaS apps.

Developed an LLM based Automated Meeting Summarizer app using Flask for internal company use which can automat-ically detect when the meetings have ended and attach a structured summary document of the meeting containing clear action items for the stakeholders involved.

Deployed the app on Google Cloud Run integrated with Google Secrets Manager to enable dynamic auto-scaling according to the workloads.

Configured and deployed an AI based ticket response agent that can classify user generated tickets and queries based on given categories and respond to them with the help of provided knowledge base documents.

Achieved a Ticket Classification Accuracy of more than 95% along with an average User Response Score of 4.1 out of 5 across different channels such as chat, voice, and email.

Configured and deployed an AI agent that can analyze the performance of customer service agents based on annotated transcripts of calls between agents and customers through LLM-based rule matching.

Achieved an Accuracy of 80% and a Rule Coverage of 95% through rigorous testing and prompt tuning techniques.

Successfully handled 3 CICD pipelines across the 20 major OCI regions worldwide as part of my role as the release coordinator.

Took full responsibility and ownership of the CI capacity transition from x86 to ARM-based machines in the Singapore region.

Built a rudimentary ML-based Incident Clustering model that can cluster the sev-1 and sev-2 incidents in an org and map them to a root cause as a helping tool for the on-call manager in all-hands meetings.

Trained the model on around 50000 Jira tickets for internal Oracle Cloud incidents and tested it using ground truth comparisons. Achieved a 90% Accuracy for RCA which could potentially save about 10-15 minutes of time taken in RCA per incident for the on-call manager.

Initiated and successfully transitioned Oracle Cloud Infrastructure Servers from the prevalent x86 to the cheaper and more energy-efficient ARM architecture machines potentially bringing down OCI server costs by 30% in the long run by importing, building, testing, and publishing the etcd binaries for internal Oracle use.

Also automated the whole process of importing, building, testing, and publishing the etcd binaries (written entirely in Golang), saving a lot of time and speeding up the transition.