About.
Software Architect at Entrata India with seven-plus years building property-management software at scale. Lately that means leading multi-agent AI inside a dense domain with a decade of legacy behind it. I care about clean architecture, developer experience, and shipping work that holds up over time.
Experience
Architect-leaning IC work across the property-management platform. Multi-agent AI systems, system design, domain-driven design, frontend modernisation, and a fair bit of mentoring.
AI and agent work
- Built a multi-agent platform for property management on AWS Bedrock AgentCore, orchestrated with LangGraph, to help residents through lease renewals across chat, SMS, and voice.
- Shipped a renewals sub-agent that answers questions on rent and charges, compares multiple offers, and lets residents accept their preferred one, plus a maintenance sub-agent that auto-generates work orders and escalates to property managers when needed.
- Connected the agents to the legacy PHP backend and database through an MCP server, with guardrails and deterministic harnesses for Entrata's dense domain SOPs.
- Owned SLA observability and drift detection with LangFuse and New Relic, and built an AWS Glue ETL pipeline that turns agent activity into client-facing effort-savings reports.
- Ran the stack on GPT-4.5 via a LiteLLM proxy, with continuous prompt layering, prompt optimisation, and MCP tool optimisation.
Delivery and architecture
- Led React adoption for key projects, cutting client-reported issues by 30%.
- Built an Amplitude SDK for local feature-flag evaluation that cut API overhead by 90% (10 calls to 1), brought decision latency under a millisecond, and scaled across multiple teams. Shipped with 100% test coverage and zero breaking changes.
- Contributed to organisational and open-source projects, including third-party adapters and performance optimisations.
Tooling and automation
- Built custom monitoring with New Relic, Papertrail, and Slack, improving incident response times by 25%.
- Automated charge-posting monitoring, saving 30 mins daily (~150 hours/year).
- Created GitHub Actions for automated workflows, increasing code review efficiency by 40% across 5 teams.
- Worked with Backslash and Cycode to bring static code analysis into our review process.
- Raised code quality through better testing habits and tighter review practices.
Mentorship and learning
- Mentored 5+ new team members, which helped lift team productivity by 25%.
- Ran training on emerging technologies and became the go-to person for new tech questions.
- Trained 25+ new joiners and multiple teams on AI prompting basics.
Debugging and reliability
- Implemented feature flag management for controlled rollouts and A/B testing.
- Used New Relic, CloudWatch, and XDebug to track down production issues and cut downtime.
Project and team leadership
- Managed critical projects with 95% on-time delivery across cross-functional teams.
- Kept quality high through active code reviews and clear design standards.
- Led migration from legacy systems to a modern service-based architecture.
- Led trigger remediation to retire complex DB functions and move them into scalable, testable, domain-shaped app-layer services without changing behaviour.
Where the obsession with shipping software that actually solves problems started.
Recognition
Caught a CI/CD silent failure where 98% of tests were getting skipped, then led the org-wide move to Amplitude Local Evaluations.
Fixed critical dependency injection issues that were blocking multiple teams from scaling.
Top rating three years running for consistent, high-impact delivery.
High-impact project delivery during a rough quarter.