About
Experience
Software Engineer II at Microsoft
Redmond
In the Office of the CTO, building and optimizing the Microsoft 365 services behind AI-powered and agentic experiences.
Software Engineer co‑op at Kyndryl
New York
Worked on the team that took a Java API gateway in front of five services from 850 ms to 280 ms at p99 and cut timeout errors 85%, with per-dependency timeouts, Resilience4j circuit breakers and query tuning.
- Rolled out JWT-based OIDC across 8 Java services, securing 1M+ API requests a month, with a staged dual-auth cutover and no downtime.
- Built an internal developer platform on Backstage that became the service-scaffolding standard for 30+ engineers.
- Automated deploys for 8 microservices across EKS, AKS and GKE with Terraform and Helm, cutting environment setup from days to under an hour.
- Java
- Resilience4j
- Spring Security
- OIDC
- Backstage
- Terraform
- Helm
- EKS
- AKS
- GKE
Software Engineer at Oracle
Bengaluru
Owned technical delivery for CX Sales (50,000+ customers) with an 8-engineer offshore team: SSR, caching and payload work that cut page load 60% and tripled release cadence.
- Shipped three NLP support bots that handle 100,000+ queries a month at 85% intent accuracy.
- Built enterprise SSO (OAuth 2.0, OIDC, SAML) across four applications, provisioning 5,000+ employees.
- Moved ten Java ETL services onto Kubernetes, processing 500 GB+ a day at 99.9% reliability.
- Mentored five new hires through code-review shadowing and pairing on production tickets.
- Java
- Next.js
- OAuth2
- SAML
- Kubernetes
- Jenkins
- Python
- Transformers
Associate Software Engineer at Oracle
Bengaluru
Built a Java migration framework that moved 250,000+ records to the cloud, with reconciliation at every cutover.
- Built five REST and SOAP connectors for Salesforce and SAP: 50,000+ transactions a day at p95 under 100 ms, and 75% less manual data entry.
- Promoted to Software Engineer after a year.
- Java
- JDBC
- REST
- SOAP
- Salesforce
- SAP
Projects
A print queue that can’t print the same job twice
2026, Northeastern, team of three
Release-at-the-printer cloud printing on AWS: a Go API, Python workers, one SQS queue per printer, and DynamoDB conditional writes instead of locks.
50 concurrent releases of one job: 1 print, 49 clean conflicts. p99 under 470 ms at 50 users.
- Go
- Python
- SQS
- DynamoDB
- ECS Fargate
- Terraform
A recommendation service for the Stock Trader demo
2025, Kyndryl, open source (Apache 2.0)
Hybrid retrieval over Azure AI Search with reciprocal rank fusion, answers from Azure OpenAI, and a Redis cache in front of both. It went from prototype to something the team used.
95% cache hit rate; cached queries under 100 ms at p99.
- FastAPI
- Azure OpenAI
- Azure AI Search
- Redis
Outpost: clinical triage that never leaves the box
2026, Dell × NVIDIA hackathon
Offline triage on a single Dell GB10 with local models for speech and medical text. The only thing that ever leaves the machine is a six-field summary a person has approved.
124 bytes of approved egress, and nothing else.
- Ollama
- MedGemma
- Whisper
- FastAPI
CodeCollab: rooms for pair programming
2024, Personal project
Developers create a room for what they want to work on, tagged by language, and others browse and join it for a live video session. Sign-in with NextAuth, rooms in PostgreSQL through Drizzle, and video calls through Stream’s SDK rather than a hand-rolled WebSocket layer.
- Next.js
- TypeScript
- PostgreSQL
- Drizzle
- NextAuth
- Stream
SceneSnap: recreate a movie scene, offline first
2025, Northeastern, team of three
An iOS app for recreating famous movie scenes and comparing your shot with the original. Posts are saved on the phone first and sync to Firestore and Firebase Storage when the connection comes back, and an engagement-ranked leaderboard updates as they land.
- SwiftUI
- Combine
- MVVM
- Firebase Auth
- Firestore
Writing
An attention window is not a checkpoint: inside vLLM’s 128-token replay
, 10 min read on Hashnode
Why vLLM sometimes recomputes tokens to save prefix-cache memory, and what that trade costs.
The model is untrusted code: a design review of Meta’s Muse
, 17 min read on Hashnode
Following one malicious email through Muse to see where a prompt-injected agent gets contained.
TOON vs JSON: the battle for token efficiency in LLM applications
, 6 min read on Hashnode
Measuring how much a data format changes what you pay per prompt.
Education
MS, Computer Science, Northeastern University
Sep 2024 – May 2026, Boston
Khoury College of Computer Sciences.
BE, Computer Science, Visvesvaraya Technological University
Aug 2017 – Aug 2021, Bengaluru
Colleagues
“From the beginning, Anand required minimal guidance and quickly became someone we could rely on to operate independently, even when goals were ambiguous or still taking shape. … Anand brought together several complex considerations into a cohesive solution that reduced our environment setup time from several days to under an hour.”
Karri Carlson-Neumann, Cloud Journey Optimization, Kyndryl. Senior colleague at Kyndryl, .Read the full recommendationHide the full recommendation
Anand was a delightful and highly impactful addition to our team. Our work is fast-paced, dynamic, and often driven by rapidly evolving requirements. From the beginning, Anand required minimal guidance and quickly became someone we could rely on to operate independently, even when goals were ambiguous or still taking shape.
During his time with us, Anand successfully drove a wide range of initiatives, including building a new recommendation service for our team’s Stock Trader demo application using Azure OpenAI, taking it from prototype to a solution the team could actively use. He also contributed important authentication capabilities across microservices and helped automate deployments across multiple public cloud environments.
The deployment automation work was particularly impactful. Anand brought together several complex considerations into a cohesive solution that reduced our environment setup time from several days to under an hour, delivering meaningful efficiency gains for the entire team.
Beyond his technical capabilities, Anand is thoughtful, collaborative, and a pleasure to work with. I was consistently impressed by his initiative, adaptability, and ability to deliver results. I strongly recommend Anand and would be excited to work with him again in the future.
“During the project Anand demonstrated not only his distinctive technical skills, becoming familiar with the new service and associated tools very fast, but also the positive attitude, team spirit, customer mindset, and leadership that was required to successfully deliver the project.”
Ronaldo Pinto, Corporate strategy and finance transformation, ex-McKinsey. Oversaw a project Anand worked on at Oracle, .Read the full recommendationHide the full recommendation
I had the pleasure of overseeing a project where Anand was one of the senior cloud analysts. The project consisted of migrating one of Oracle’s largest CX cloud customers to an innovative service, under a very constrained timeline. During the project Anand demonstrated not only his distinctive technical skills, becoming familiar with the new service and associated tools very fast, but also the positive attitude, team spirit, customer mindset, and leadership that was required to successfully deliver the project. I would jump at the opportunity of having Anand in my team again.
“Anand is very technically adept but more so looks for better ways and options to do things.”
Nathan Turner, Director Technical Manager. Worked with Anand on several Oracle projects, .Read the full recommendationHide the full recommendation
I have worked with Anand on several Oracle projects and would welcome the opportunity to have Anand on future projects. Anand is very technically adept but more so looks for better ways and options to do things. Anand has been great to work with on every project.