I'm the Founding Forward Deployed Engineer (engineer #3) at MoolAI. I build the platform that takes enterprise AI agents from pilot to production with deterministic guardrails, and I deliver those agents for paying customers in EPM, healthcare and network security.
Before MoolAI I was a Data Engineer at LTIMindtree, running PySpark, Kafka and Databricks pipelines at the terabyte scale. I hold an MS in Analytics from San Francisco State University.
| Before | After | How |
|---|---|---|
| 4-day EPM financial review | 30 minutes | A hierarchical multi-agent system: a supervisor that plans and delegates to retrieval, anomaly-flagging and narrative sub-agents |
| 7-day logistics reconciliation | 1 day | A multi-agent pipeline that replaced manual coordination across disconnected supply-chain systems |
| Manual security alert triage | 50% lower latency, 99.9% uptime | A ReAct-style LangGraph + MCP system deployed in an air-gapped, on-prem environment (I led a team of 3) |
| Per-seat licenses for EPM, ERP and accounting tools | Usage-based access from Slack and Teams | Conversational agents in production at 3 enterprise customers |
- Judge Council (provisional patent filed): a multi-stage LLM-as-a-judge evaluation pipeline. Judges score independently, then it calibrates across judges, corrects for verbosity bias and combines the results into a consensus. It runs in the Experiment Engine and against sampled production outputs.
- ContextForge (provisional patent filed): a declarative RAG/CAG context-engineering engine. It compiles context in five phases, and when validation fails it feeds the schema and runtime errors back into the prompt so the agent can try again.
- Agent deployment plane: turns configured agents into signed artifacts and reproducible Terraform deployments. It works in customer-tenant Azure and MoolAI-managed Azure, gives one-click pilot-to-production promotion and keeps staging and prod infrastructure identical.
MoolAI · Founding Forward Deployed Engineer · San Francisco, CA · May 2025 – present Platform architecture and end-to-end enterprise delivery, from technical scoping after the sale through production go-live. I'm the technical point of contact for customer leadership.
LTIMindtree · Data Engineer · Bengaluru, India · Aug 2021 – Apr 2023
- Processed 2TB+ of IoT data daily with PySpark streaming, cutting end-to-end latency by 40%.
- Built Delta Lakehouse pipelines in Databricks that unified ERP and cloud data, reducing ETL from 7 days to 2.
- Built a PySpark + Kafka ingestion framework that auto-registered 10,000+ records per cycle in Alation.
- Set up Azure DevOps CI/CD, raising deployment frequency by 40% with zero-downtime releases.
Decentralized edge LLM deployment for private GenAI · Graduate research, SFSU (advisor: Prof. Sameer Verma) Private LLM inference on a Raspberry Pi cluster using MicroK8s, Ollama and Distributed Llama. On-device RAG with no cloud dependency achieved under 150 ms responses, 35% lower latency than the baseline and 25% lower power use.
Languages
Agents & GenAI
Cloud & infrastructure
Backend & data
Enterprise integrations: Slack API, Microsoft Teams API, Workday Adaptive Planning, Pigment, SAP, Salesforce, QuickBooks
- 🎓 MS, Analytics, San Francisco State University (2023 – 2025)
- 🎓 BE, Computer Science, Visvesvaraya Technological University (2017 – 2021)
- 📜 Google Professional Data Engineer · AWS Solutions Architect – Associate · Microsoft Technology Associate (Python)
Always happy to talk about agent architectures, LLM evaluation or getting GenAI into production. Reach out on LinkedIn.