Staff Software Technician · Build, Test & Release · Bengaluru

I build the tooling that explains why builds break.

I work on the Build, Test & Release team at NI (Emerson), where I lead an LLM-agent system that reads failed builds so engineers don’t have to. I’m also studying AI & ML at BITS Pilani, and I’m as interested in applied ML as I am in the build and release work that got me here.

Read the case studies Resume (PDF) Email me

Failed builds triaged
1,350+
Resolved with no LLM call
~65%
Eval cases passing
217/250
Saved per failure
10-15 min
Linux feeds published
200+

Build Failure Analyzer

Role
Built it, now lead
Since
2024
Status
In production
Used by
Teams across NI

When a CI build fails, someone has to read the logs, decide whether the cause is the infrastructure or a code change, and work out which pull request broke it. That happens dozens of times a week. I built a system that does the first pass.

How it works

about 65 in 100 settled by fixed checks, no model call about 35 go on to the agents
  1. A failed build lands in the review queue.
  2. Deterministic checks run first: known infrastructure patterns, duplicate detection, failure clustering and dependency tracing. about 65% of failures end here
  3. What’s left goes to a root-cause agent. It reads the logs and PR history through tools and names the pull request that most likely caused the break.
  4. A second agent, running on a different model, checks that call before anything is posted.
  5. The result is commented on the work item. A triager confirms it, and only then does the system file or link a bug.

Models are slow, cost money and are sometimes confidently wrong. So the cheap checks go first, a second model reviews the first, and a person signs off before anything is filed. Different jobs use different models: Claude Sonnet for root cause, GPT as the validator, Claude Haiku for clustering.

Results

  • 1,350+ failed builds analyzed across 600+ work items.
  • Saves roughly 10-15 minutes of manual triage per failure.
  • Scores 217 of 250 on a labeled eval set, which I use to check every change to the prompts and rules.
  • Won an R&D Excellence Award in Emerson’s 2026 R&D Global Awards.

Where it falls short

It still gets 33 of my 250 eval cases wrong, and a model can be wrong with total confidence. That’s why a triager’s sign-off is a hard gate before any bug is filed, not an optional step.

Stack Python · Azure DevOps REST API · GitHub Copilot SDK · SQLite · pytest · Poetry

Infrastructure Failure Review Agent

Role
Built it
Since
2026
Status
In review
Ships as
Python package + agent skill

An AI reviewer reads the build-failure report and writes up the infrastructure problems. The trouble with leaving that to a model is that two runs on the same report can disagree. I made it predictable.

  • The reviewer must report every infrastructure cause that hits five or more distinct builds. That rule lives in code, not in the prompt.
  • It can still add rarer failures it thinks are worth a look, so judgment stays where it helps.
  • Each finding traces back to the builds behind it, so a reviewer can audit it.
  • Ships as a tested Python package with an agent skill and its own Azure DevOps pipeline.

Stack Python · LLM agent skill · Azure DevOps Pipelines · pytest

Linux feed publishing pipeline

Role
Owner
Since
2024
Status
In production
Used by
Teams across NI

Publishing a Linux package feed used to be a long manual checklist. I turned it into an Azure DevOps pipeline that handles the whole run, including the CDN cache purge and SHA256 checksum checks. It is now how NI publishes all of its Linux feeds.

  • Used across the company to publish every Linux feed.
  • Published 200+ feeds through it.
  • Saves 3-4 manual hours on every patch or feed release.
  • The sync_feeds pipeline is also used by NI’s Penang site.

Stack Azure DevOps Pipelines · Python · YAML · DEB/RPM/IPK packaging

Smaller things

Rahul Bandi laughing, wearing glasses and a black shirt, seated outdoors

I started with a diploma in computer engineering and information technology at NTTF. Along the way I competed in the IEEE Brixia European CanSat and DSU CanSat competitions with a small satellite that sent sensor and GPS data over radio.

In July 2024 I joined NI’s Build, Test & Release team, where the job is keeping the CI that ships hardware drivers healthy. I like the debugging: reading a failure, working out what actually broke, and then making sure nobody has to do that by hand a second time. I became a Staff Software Technician in October 2026.

Now I’m studying AI & ML at BITS Pilani while working full-time, and the Build Failure Analyzer is where the two meet.

How I like to build

  1. Run the cheap checks first. About 65% of failures are settled before any model is called.
  2. Measure before changing anything. A 250-case eval set tells me whether a prompt change helped.
  3. Keep a person in the loop. A triager confirms every call before a bug is filed.
Based in
Bengaluru, India
Studying
B.Tech, AI & ML, BITS Pilani (work-integrated, since Jul 2025)
Diploma
Computer Engineering and Information Technology, NTTF, 2021 - 2024 (grade 9.07/10)
Learning
Mathematics for ML, discrete maths and graph theory, advanced Python, system design

National Instruments (Emerson) Bengaluru, India

Oct 2026 - now

Staff Software Technician, Build, Test & Release

  • Lead the Build Failure Analyzer, the multi-agent system NI uses to triage failed builds.
  • Own build and release for 65+ drivers within a 27-person team.
  • Own the Linux feed publishing pipeline NI uses company-wide, saving 3-4 manual hours per release.
  • Mentor newer teammates on investigating CI failures.

Jul 2024 - Sep 2026

Software Technician, Build, Test & Release

  • Built the first version of the Build Failure Analyzer.
  • Wrote Python and Selenium tools that replaced manual dashboard checks and flagged stalled CI events, saving about 350 engineer-hours per release cycle.
  • Published 200+ Linux package feeds through Azure DevOps.
  • Tracked down CI failures from a flaky FlashCache unit test to an SBOM artifact name collision and a build-tool EOF bug.

If you’re hiring for ML engineering, build and release or DevOps, or you just want to talk about CI, write to me.

rahulbandi568@gmail.com