I lead the teams that ship production LLM systems — agent orchestration, cost engineering, and HIPAA-compliant SaaS at scale. 14+ years, hands-on the whole way.
// how a Jira ticket becomes a merged PR — the SDLC platform I built at ShyftLabs
Production-grade platforms across US healthcare SaaS, autonomous AI agents, biometric security, FinTech, and TaxTech — all designed for scale, compliance, and measurable outcomes.
A two-pass agent pipeline that classifies Jira tickets, implements fixes, and gates risky changes behind Slack approval — zero manual triage.
Organization-level guardrails that protect secrets, block dangerous operations, and enforce managed AI coding controls across developer platforms.
Custom Claude Code skills that analyze and classify production logs and Jira history, then send daily Slack reports with constructive, consolidated comments on recurring, reproducible, and newly generated bugs.
An 18-step content DAG generating meta tags, copy, FAQs and schema across 300+ healthcare websites, with multi-model fallback and cost guards.
Architected a fully isolated multi-tenant SaaS platform hosting 500+ US healthcare websites, each with per-tenant configuration, branding, data isolation, and independent scaling — with zero cross-tenant data leakage.
Delivered a HIPAA-compliant Booking Management System with live slot availability sync, real-time PIMS (Practice Information Management System) integration, encrypted PII data handling, and full audit trail — increasing partner adoption by 45%.
A face authentication platform for SaaS and on-prem deployment, with a biometric model that keeps learning a user's face over time.
Led end-to-end product strategy for large-scale TaxTech SaaS. Engineered a dynamic tax determination engine with 50+ rule-based algorithms, automating complex TDS computations with 99.9% accuracy.
Designed a HIPAA-compliant healthcare product for European startups. Implemented secure patient data exchange and remote consultation features, ensuring strict compliance with health data privacy regulations.
Architected a high-traffic news portal on AWS (EC2, S3, RDS, CloudFront) handling 50K+ hits/hour with automated RSS feed ingestion and WordPress-S3 synchronization.
Delivered FinTech mobile banking solutions for 165+ cooperative banks using C# and Flutter, ensuring 99.99% uptime and 1M+ secure transactions/month per bank.
Launched a scalable mobile wallet with 100K+ active users, integrated BillDesk payment gateway, enabling instant UPI and bill payments with <2 sec transaction latency.
Partnered with World Bank on geofencing and NO₂ emission monitoring, leveraging Python, PostgreSQL, and geospatial analytics for real-time air quality visualization.
Co-founded and built a unified mobile banking app in Java, enabling 70+ banks to perform secure, real-time view-only operations — reducing support load by 40%.
LLM pipelines, Claude Code skills, agent orchestration, cost & token engineering, RAG and multi-model fallback.
Leading 10–20+ engineers through delivery, hiring, and agile execution.
TOGAF-aligned enterprise architecture, AWS-native platforms, multi-tenant SaaS and microservices.
Turning business goals into shipped roadmaps — on time, at 98%+ delivery rates.
I gravitate toward systems where the constraints are real — regulatory (HIPAA, ISO 27001, GDPR, RBI), operational (300+ tenants, 10M+ records a day), or architectural (agents that have to be trusted with production code). The common thread has been owning it end to end: architecture, team, and delivery.
As an Engineering Manager I'm accountable for a roadmap and a team of 10–20+ people, but I still design the systems and write meaningful amounts of code — the SDLC agent pipeline, the biometric matching engine, and the tax rule engine were all things I architected personally.
Most interesting problems aren't about the model — they're about cost, safety, and control. Bounded agent turns, Slack approval gates, multi-layer caching, and per-tenant cost guards. I'd rather ship a system that's slightly less magical and fully explainable than one that's impressive in a demo and unaccountable in production.
Field notes on agentic AI, LLM cost control, enterprise AI security, and healthcare platform architecture—grounded in the systems I have built.
How bounded agent runs and risk gates make autonomous SDLC workflows safer to operate.
Read insight →A practical framework for caching, quality gates, fallback models, and spend controls.
Read insight →Managed policies, permission boundaries, and platform-aware safeguards for coding agents.
Read insight →How tenant isolation, auditability, and access control shape reliable regulated platforms.
Read insight →Open to engineering leadership roles, AI advisory, and conversations about production LLM systems.
Based in Noida, India · open to remote & relocation conversations