// Senior Software Engineer

Shreyas
Kaushik

Building agentic AI platforms and cloud-native infrastructure at scale. At Mphasis Applied AI I build ATOM, an on-premises agentic development platform, and AEGIS — the governance & information-security control plane securing 500+ agents that serve 1M+ requests a day inside Mphasis.

500+
agents in production
1M+
agent requests / day
10M+
messages/day systems

A bit about me

I'm a Senior Software Engineer who specialises in agentic AI systems and cloud-native infrastructure. Started my career at IHX processing millions of medical documents daily, where I learned what it means to build systems that can't go down. Now at Mphasis Applied AI, I build ATOM — an on-premises agentic platform — and AEGIS, the governance and information-security control plane around it. Together they run 500+ agents handling over a million requests a day inside Mphasis.

I care most about the control plane around agents — auth, policy enforcement, observability, isolation. The agent logic is often the easy part. The infrastructure that makes it safe and auditable is where the real engineering is.

key highlights
Built ATOM + AEGIS — an on-prem agentic platform and its governance layer. 500+ agents in production serving 1M+ requests/day internally, every LLM call policy-checked and immutably logged.
Reduced DB query time from 1–2s to 10–15ms across 10 tables at IHX.
Status notification system processing 10M+ messages per day using SNS.
Document intelligence platform that improved first-submission claim rate from 23% → 77% in 2 weeks. Now classifies 4M+ images/month.

Where I've worked

MPHASIS
Senior Software Engineer, Applied AI
Feb 2025 → Present
ATOM — Agentic Development & Deployment Platform
AgentScope · TypeScript · Golang · FastAPI · Docker · Kubernetes · Kafka/Redpanda · PostgreSQL · Redis · Grafana Alloy · Loki · Tempo
On-premises agentic platform used internally across Mphasis — engineers write agents via SDK and deploy with a single command, while infrastructure, LLM routing, and lifecycle are fully abstracted away. Runs 500+ agents in production serving over 1M requests/day. Agents execute as isolated containers via atom-runtime; Studio handles lifecycle, observability, and human-in-the-loop.
AEGIS — Agentic Enterprise Governance & Information Security
LiteLLM · FastAPI · Python · Golang · OPA · PostgreSQL · Redis · Docker
The governance and security control plane for every agent on the platform. Each LLM call flows through AEGIS for JWT auth, OPA policy enforcement, rate limiting, PII-aware routing, per-tenant budget governance, and an immutable audit log — with compliance coverage across PCI-DSS, FIPS, GDPR, and DORA.
IHX — A PERFIOS COMPANY
Senior Software Engineer
Jul 2024 → Feb 2025
Orchestration Engine
Netflix Conductor · Zookeeper · Redis Sentinel · TypeScript
Comprehensive workflow orchestration for document processing pipelines — extraction, OCR, classification, quality control. Parallel processing patterns reduced processing time by 40%.
Persistence Engine
TypeScript · TypeORM · PostgreSQL · Redis
Single-layer persistence solution reducing average query time from 1–2s to 10–15ms across 10 tables.
Status Notification Service
TypeScript · AWS SQS · AWS SNS · MongoDB
High-throughput system processing 10M+ messages daily, providing real-time status visibility across AI/ML entities to client dashboards.
IHX — A PERFIOS COMPANY
Software Engineer
Aug 2022 → Jun 2024
Document Intelligence Platform
TypeScript · FastAPI · AWS EKS · MongoDB · S3
Document processing platform for India's largest health benefits administrator. Took first-submission claim rate from 23% → 77% in 2 weeks. The platform now classifies over 4 million medical images every month.
Infrastructure Transformation
Kubernetes · Helm · NGINX · Grafana · Loki · Promtail · Prometheus · Service Accounts
Led Kubernetes and AWS infrastructure design for AI model security and scaling. Introduced full observability stack, enabling rapid feature deployment.
Authentication Solution
Keycloak · Java · OAuth 2.0 · JWT · NGINX
MFA and SSO with Keycloak custom plugins across multiple services — reduced login time by 50%.

Tech stack

Languages
TypeScript Python Golang JavaScript Bash SQL
Agentic AI & LLMs
LLM Gateway design Multi-agent orchestration Tool use RAG pipelines OpenAI API Anthropic API Gemini API
Web Technologies
Node.js FastAPI Express React.js Django NGINX
Cloud & DevOps
AWS EKS AWS S3/SQS/SNS Docker Kubernetes Helm GitHub Actions
Databases & Messaging
PostgreSQL MongoDB Redis Redis Sentinel Kafka / Redpanda TypeORM
Observability & Security
Grafana Loki Promtail Prometheus OPA Keycloak OAuth 2.0 JWT

Things I've built

Technical writing

Get in touch

Open to conversations about interesting engineering problems, agentic AI infrastructure, and roles where the work involves real systems at real scale. Not looking for something passive — if you have a hard problem, let's talk.

location
Bengaluru, India
$ cat contact.json
{
  "status": "open to conversations",
  "interests": [
    "agentic infra",
    "high-scale systems",
    "ai security"
  ],
  "response_time": "fast"
}