00/12 · COVER · BLACK + TEAL + SUNFLOWER
Sarosh Aga2017
Selected work
2017 → now
Sarosh Aga · engineering, leadership, applied AI
0%
·······
Sarosh Aga
2026hello
Engineering Manager · Applied ML and GenAI · Dubai

SaroshAga.

I lead engineering teams building applied AI for the open web.

Nine years, from platform engineering to GenAI. Every scene below is drawn live as you scroll.

Sarosh Aga · Dubai
scroll through the history
0a6f2e42017 · Platform engineer

Data at industrial scale.

Associate → Software engineer → Senior · 2017–2020

Three years on an enterprise ad platform, building the horizontal services behind ad buying, selling and campaigns.

  • Promoted twice in three years, associate to senior.
  • Built a master data REST framework and a way to standardise data definitions across apps.
  • Developed an ETL framework ingesting digital and print ad data: characteristics, financials and performance.
  • Maintained document delivery, locale management and distributed caching services.
JavaSpringOracleETLDistributed caching
print + digital → one schema
e7a0d152020 · Hosting platform

Enterprise out, consumer in.

Senior software engineer · 2020–2022

A move to Dubai, and to the consumer web: a hosting platform behind a very large number of websites.

  • Migrated developer and self-serve tooling from a custom hosting orchestrator to Kubernetes, improving developer experience and performance.
  • Built a site-details ingestion pipeline, data index and API that cut ticket response times and enabled proactive health checks.
fig. 1 · New York office, in person
KubernetesDeveloper toolingData pipelines
moving tooling to k8s
1c8e4b22022 · Engineering manager

Taking the helm.

Software engineering manager · 2022–now

One squad, several mandates, and plenty of people coming and going along the way. The constant was how the team was run.

  • Leads a squad of up to nine engineers, owning hiring, onboarding, performance reviews and career development.
  • Owns technical strategy and roadmap end to end for select initiatives, and co-owns multi-quarter strategy with product and senior leadership.
  • Directs resourcing and prioritisation across new builds, maintenance and operations running in parallel.
  • Kept the squad productive through team transitions: onboarding, role changes and extended absences.
  • Instituted cross-team documentation standards: RFCs, product specs and tech specs.
Up to 9engineers
4 yrsmanaging
hiring
9b2d7f3Mandate · Data and analytics

Listening to the noise.

Engineering manager · data and analytics mandate

The squad’s mandate shifted to data: knowing how every customer application was performing, and noticing early when it wasn’t.

  • Built a data warehouse and ETL pipeline for application performance and health, with a REST API and charting framework.
  • Gave the platform baseline views: response times, cache, queries, CPU, memory and PHP threads.
  • Designed an anomaly-detection library of statistical daemons with configurable thresholds and alerting.
  • Cut time-to-detection for DDoS attacks, memory leaks and traffic spikes, from election nights to Super Bowl Sunday.
Data warehousingETLAnomaly detection
requests / sec
c51e08aMandate · Data and analytics

Let the bandits pick.

Headline testing with Thompson Sampling

Customer interviews showed editors wanted to test many headlines at once, not run A/B tests.

  • Built a headline tool on a multi-armed bandit with Thompson Sampling.
  • Major newsrooms optimised breaking-news headlines in ~15 minutes.
  • Grew the product’s ARR by 120% in four months.
~15 minto a winning headline
+120%ARR in four months
real Thompson Sampling
3d4a6b1Mandate · Applied ML and GenAI

Agents that build websites.

Engineering manager · applied ML and GenAI mandate

The latest mandate: customer-facing AI on the company’s largest product, built on an in-house agentic framework.

  • Leads the squad building the agentic framework behind the company’s LLM-backed apps.
  • Built an agentic site-specification workflow that, chained with other agents, became the AI site builder.
  • Reached ~2,000 daily active users.
~2,000people a day
reading briefagentic
8f27c3dGenAI · Images

Noise in, pictures out.

Generative image editing

Image generation and editing, right where people write.

  • Shipped the AI image editing and generation tool, at 1,000+ daily active users.
  • Investigated a 27% dip after launch: overall editor usage had fallen, while the tool’s conversion held steady.
1,000+people a day
›
b6e91f0GenAI · In the editor

Every post, a short film.

Collaborative writing, video and voice

AI inside the editor, from first draft to finished video.

  • Built a collaborative AI writing experience in the editor.
  • Built AI video that turns any post into a 9:16 vertical video in about a minute, composing the headline, visuals and a closing call to action.
  • Contributed to site-wide content guidelines: a store of a site’s voice, tone and terminology that keeps AI writing on-brand.
blog post
4e1c9a02026 · After hours

The city, as data.

Side project

Interactive maps of Dubai’s district-cooling demand and property prices.

  • Built map explorers for cooling demand and property prices, rents and off-plan supply.
  • Adding time-series forecasting next.
fig. 2 · Abu Dhabi, last light
cooling demandDubai
In summary

Nine years,
four threads.

Writing and talks

Notes from
the work.

    Speaker and panelist

    Dubai Software Engineering Management Group, 39th meetup

    Attended

    LeadDev New York, with a write-up

    Contact

    Let’s build
    something.

    © Sarosh Aga · Dubai · Drawn live, two inks per chapter.