Skip to content

Ask me about him. I'll answer and take you there.

Ask your own question

Thatikonda Mukesh / AndysTMCIndia

Mukesh

I make software work the way it is meant to, and keep it working.

Software engineer. Full-stack, cloud and AI systems that I build and run myself.

Right nowBuilding Teamora, a real-time workspace on a Kubernetes cluster I built, and Mei, an AI assistant for the Linux desktop.

Looking forFull-stack, AI or DevOps roles. Hyderabad, Bangalore, Chennai or remote. Can join immediately.

Based in IndiaGet in touch →

What have you built?

NowTeamora5 services · my own Kubernetes clusterA whole workspace, on a cluster I built.

Documents, sheets, slides, whiteboards, tasks, chat and meetings in real time, as five Python microservices. It runs on a Kubernetes cluster on AWS that scales from zero using an operator I wrote in Go. Its assistant answers only from cited quotes and pauses for approval before acting.

Read the full story →

Teamora screenshot — Demo ▸

GitTypeScriptHTML and CSSAutomated testingReactRESTAWSDocker+42 more

NowMeiAbout 600 automated testsAn AI assistant that lives in the desktop.

A GNOME assistant that can work with files, apps, settings and commands, but only with your approval. The approval rule is enforced outside the model, so a malicious web page cannot talk its way past it.

Read the full story →

Mei screenshot — Demo ▸

GitTypeScriptAutomated testingLLM APIsPythonArchitecture decision recordsLinuxTool calling+12 more

2026patrn.inkEvery change in a signed public logA link shortener you can verify.

First built in Go with my brother, then rebuilt in Java. Every destination change is written to a signed transparency log, so anyone can check a link was not quietly redirected. Analytics store no IP addresses.

Read the full story →

patrn.ink screenshot — Demo ▸

GitTypeScriptHTML and CSSAutomated testingReactJavaRESTAWS+29 more

2026Demo platformScales from zero · no open portsServers that sleep until someone visits.

On-demand hosting for my demos. Each app stays off until a visitor presses Launch, runs a short session, then stops itself. A Kubernetes operator I wrote in Go starts and drains the machines.

Read the full story →

Demo platform screenshot

GitAWSKubernetesOIDC federationLinuxGoDynamoDBPasskeys (WebAuthn)+4 more

2024Track Your Attendance3,000+ studentsThe app a third of my university used.

Timetable and attendance for SRM-AP students: live class countdowns, how many classes you can skip, and how many you need to reach 75 percent.

Read the full story →

Track Your Attendance screenshot

GitTypeScriptHTML and CSSNext.jsTailwind CSSMongoDBJavaScriptFramer Motion+5 more

NowUniNeeds7 services · serving live trafficA new backend for a live app that could not change.

A college services app built by my teammate Gowrish. I built and deployed its whole backend: a Spring Boot gateway and seven services that now serve the app's live traffic, with no change to the app.

Read the full story →

GitAutomated testingJavaRESTDockerSQLSpring BootMicroservices+16 more

NowLittle LaterNothing leaves your machineSave it now, deal with it later.

A Chrome extension with no backend. Bookmarks, tasks, reminders and notes stay encrypted on your device, and the AI helper will not write anything until you confirm.

Read the full story →

Little Later screenshot — Demo ▸

GitTypeScriptHTML and CSSAutomated testingReactLLM APIsGitHub ActionsTool calling+7 more

2025andystmc.meYou are looking at itThis site.

A one-page portfolio that answers what people usually ask about me, then takes their own questions. It replaced an earlier, database-backed site: everything here is typed data taken from my résumé notes.

Read the full story →

andystmc.me screenshot

GitTypeScriptHTML and CSSAutomated testingReactLLM APIsNext.jsGrounded answers+1 more

What do you actually know?

Not a list of claims. The bigger the word, the more places I have used it. Pick one or several to see where.

Languages

Backend and full-stack

Cloud, data and infrastructure

AI

Security and quality

Operations and production

Pick one or moreNothing selected

Where

Choose a skill above. Choose several to see where I used them together.

How do you work?

For over a year at Cognizant I kept five of Macy's financial applications running in production: more than 100 incidents and about 50 work requests.

When production breaks

The Tuesday batch that failed

Problem: A weekly batch that loads financial data for Macy's distribution centres failed one Tuesday. Nobody had changed the code.
What I found: The file server had been moved from SMB1 to SMB2. The application's Java code still spoke SMB1.
Result: I had compatibility restored so processing could resume, then changed the Java code to support SMB2 and handed it to development. A code fix, not only an escalation.

The emails that reached the wrong people

Problem: An application was sending sensitive notifications to users who should have been inactive.
What I found: Not the screen, the rules or the queries. I followed the data upstream: our copy was stale because the sync job could not reach its source through a network restriction.
Result: I raised the access request with the evidence, watched the next run, and confirmed the fix with the business users. The symptom was several layers from the cause.

6 more cases

Duplicate records that kept breaking a load

Problem: Cost data shown to business users was incomplete. A SQL Server load was failing part-way whenever the upstream system sent duplicate transactions.
What I found: I first corrected the data with SQL. When it kept recurring, I worked with a developer through the load logic and we traced it to how records were grouped before insertion.
Result: We changed the grouping so duplicates are handled before the insert. From a repeated workaround to a permanent fix.

A nightly job waiting for someone to press Yes

Problem: Data was missing in a downstream application, and its team asked whether we had sent it.
What I found: Our data was there. I followed it stage by stage and found a scheduled job stopped at a Yes/No prompt that an unattended job cannot answer. I ran the command by hand to see it.
Result: I added the parameter that skips the confirmation, and the pipeline completed.

A restart is not a recovery

Problem: Services had stayed stopped on a server, so a weekly batch across about a dozen microservices never ran. Restarting them did not help: the system had already moved on to the next batch.
What I found: I worked out how batch dates and states were kept across several MongoDB collections and in the files the jobs generate.
Result: We realigned the missed records with the current batch, rehearsed the changes on a copy first, then applied them in production. The data reached every downstream system.

Waiting on someone else's API

Problem: Some records stopped moving through the pipeline and downstream data was incomplete.
What I found: I traced them to a stage that needs another team's API, called it directly, and found it unreachable. Their team confirmed an outage.
Result: We held the jobs so nothing incomplete went downstream. When the API returned a day later, the records continued on their own with no manual fixes.

The site that loaded but showed nothing

Problem: Users could open the application but it stayed on a loading screen.
What I found: The browser's developer tools showed server errors on some requests. Code, services and database all checked out, so I widened the search to the network.
Result: A network security policy had closed ports the servers needed. I gave the network team the addresses and ports, and access was restored.

Keeping work going while a module was down

Problem: After an upgrade, a module people used to manage project records stopped working, and rolling back did not fix it.
What I found: I read the code to learn how the module normally stores its data.
Result: For each request I ran SQL to make the change and verified every result myself, so the work continued until developers restored the module.

I also automate my own work: a log analyser I built cut a 30 to 60 minute investigation to a few minutes.

By habit

The AI asks before it acts.

Mei: approval enforced outside the model
Teamora: actions pause for the user
Little Later: confirm before any write

If it ships, it is tested.

Mei: about 600 tests
Teamora: 171 + 92 tests, 10 browser flows
patrn.ink: 122 tests in release CI
UniNeeds: 1,373 tests, 826 on real databases

Decisions get written down.

Mei: 38 design decisions
Teamora: 17 architecture decisions

Where have you been?

2026AWS Certified, twiceCloud Practitioner (925 / 1000) and AI PractitionerVerify Cloud Practitioner · 597e89ad8e974dd79ad5c57f7d5b3996 ↗Verify AI Practitioner · 8598e94694a04b87b8ac9c19ccd43452 ↗

Jul 2025 – Aug 2026Jr Software EngineerCognizant · Macy's financial applications

Mar – Jun 2025InternCognizant · a team ASP.NET Core app, C# and .NET training

2021 – 2025B.Tech, Computer ScienceSRM University, Andhra Pradesh · CGPA 8.19

Along the wayCompetitive programmingLeetCode 1,793 (top 10%) · CodeChef 1,607 (3-star)LeetCode profile ↗CodeChef profile ↗

Yours.

Those were the four questions everyone asks. Ask your own: the assistant answers from my real record and shows you the proof.

Pick a question, or ask your own.

Or talk to the real one

CV (PDF)Email GitHub LinkedIn X LeetCode CodeChef