Working at Alexa - a dream come true

I happen to be one of the lucky few who have had the opportunity to work in alexa - the voice assistant that most people use for just playing music and setting alarm. But in the process I have learnt so much, not just about software engineering but also about Engineering and Operational Excellence, being an effective leader and making good decisions.

Alexa SCI - first and last line of defence.

Now that Alexa+ has launched, I can at least mention about “classic alexa”, because even back then, the LLM driven flows were being talked about. And we were handling the entire content moderation internationally for alexa. And our job was to be the first and last line of defence.

In alexa sci, I was part of several initiatives, mostly in MLOps side of things, dataset generation, orchestration, mitigation and inference. Whenever you speak to alexa, our code scans that, checks for violations and takes corrective action.

The same thing happens during response. The response is also scanned for violations. The whole thing had ML, rules, heuristics and everything a software engineer dreams of!

I did contribute a lot in several initiatives, both internal and external. From annotations tooling to developing applications for custom browsers. From python, javascript to java and even a little bit of ML, I had to dabble in everything.

The single biggest moment I had in this team was when we had to add encryption in transit to our entire application stack (that runs on a browser without debugging tools) and an internal deep dive and the latency had increased from 10s all the way to 60s. A cross team technical deep dive resulting in an optimization not only for our team (back to 10s - negligible encryption overhead), but across the whole amazon, resulting in an impact of several millions of dollars.

Pioneering work at revamping a critical alexa team

During the layoffs, a significant team inside of alexa, with touchpoint beyond just alexa, was among those impacted. And we took ownership of the team. During that time, we (me and a senior) had taken over and started exploratory work, with little documentation or guidance (as the impacted team was laid off). The team’s work included direct touchpoints within the alexa ecosystem, voice interaction design, and it served both internal and external (alexa skill owners).

Thereafter, we brought the services within operational capacity, a team within india got hired and ramped up to extend the feature set and I am still within those lovely peers.

In this team, at times, we often had solved 60+ bugs in a single day. We also was responsible for ramping up a whole team who all are still doing excellent work.

Alexa LLM efforts

We were working closely on LLM efforts as some of our services needed to be updated to support the LLM efforts and we were exploring how might we leverage LLMs to improve the efficiency of certain functionalities within alexa.

While we didn’t directly build the LLM stack, we worked closely with the teams for LLM related work. And in that time I learnt a whole lot about LLMs, some mandated, some out of my own curiosity.

I quit alexa

What my family viewed as a career suicide, I still beg to differ. My love for building the next generation of interactive experiences is something perhaps best realized independently. I am blessed to have had incredibly amazing mentors, supportive managers and a privileged position of working in a product I considered to be “sci fi” when it initially came out.