I’m Ankit Prakash, a
Eleven years shipping software at Atlassian, Oracle and Razorpay — now building with AI.
two roads
They ran alongside each other for longer than I realised. Choose the one you’d rather walk.
the shape of it
From a metallurgy degree to legal data to incident intelligence — where each road started, and what I was by the time I left.
swipe the timeline — it opens on where I am now
track one · the builder
Ankit has spent his career on the surfaces enterprises actually touch — database consoles, payment platforms, incident dashboards. The kind of work that is invisible when it goes right.
2026 — present
Atlassian · Bangalore
Ankit works on IT Service Management — the platform enterprises reach for when something breaks. Incidents, changes, service requests, and the status pages the rest of the world watches during an outage.
“A status page is the calmest surface of the loudest moment. I like that the work is invisible when it goes right.”
2020 — 2026
Oracle · Bangalore
One of the first developers Oracle hired in India for its Cloud Infrastructure team. He arrived as employee-one on the database console and left running the team that owns it — promoted from Senior Engineer to Manager in December 2023.
“I joined as the first frontend engineer for the console in India and left managing the team that owns it. That arc is the thing I’m proudest of.”
2018 — 2020
Razorpay · Bangalore
Fintech at bank scale. Ankit architected PoshVine, the rewards and offers platform behind Axis, ICICI, HDFC, SBI and Jana Small Finance Bank, and led the engineering team through the company’s hypergrowth years.
2016 — 2018
Healthi · Bangalore
Healthcare SaaS, and the first product Ankit owned end to end. He led development of the web application and designed Smart Select — an intelligent questionnaire that recommends the right health package from how someone answers.
2014 — 2016
Thomson Reuters · Bangalore
Where it started. Ankit worked on the digital transformation of legal data on FindLaw, making case law and legal information something attorneys could actually search rather than merely store.
track two · the scientist
The same eleven years, read through a different lens. Every job I have had was about inferring what someone needed before they finished asking — the models just made that tractable.
2024 · the turn
Deakin University · Melbourne
Eleven years in, Ankit went back to school — not to leave engineering, but because the interesting problems had started moving somewhere his existing tools could not reach.
“I didn’t want to be the engineer who wires up someone else’s model and hopes. I wanted to know why it works.”
2026 · Atlassian
LLMs in production
An LLM feature that reads incident playbooks and summarises them automatically, then suggests the right stakeholders by learning from historical patterns and feature ownership. It shortens the worst part of an outage: working out who to wake up.
2026 · Atlassian
multi-step prompting & tool orchestration
Agents wired into pull-request comment analysis that propose review suggestions rather than waiting to be asked. Ankit also authored the SKILL.md documentation framework that standardises how code agents plug into development workflow — the reason adoption spread across engineering teams rather than stalling in one.
2026 · Atlassian
reliability engineering for language models
Tuned system prompts and few-shot examples for ITSM-specific tasks and drove the hallucination rate down by most of its original magnitude. Treating prompts as an artefact with a regression suite, rather than a lucky sentence.
2020 — 2026 · Oracle
natural language over databases
LLM-powered natural language queries that let customers navigate a database schema by describing what they wanted instead of learning where it lived. Support queries fell noticeably; time-to-productivity improved for people who had never opened a console before.
2020 — 2026 · Oracle
instrumentation that argues back
Adobe Analytics wired into an insights engine that automated feature usage and funnel analysis — turning a dashboard nobody read into product decisions the team could defend.
2018 — 2020 · Razorpay
collaborative filtering & behavioural clustering
Bank offer inventories are enormous and almost entirely irrelevant to any one customer. Ankit built collaborative filtering with behaviour-based clustering to fix that — engagement with offers and wallet activation both moved sharply.
2016 — 2018 · Healthi
the first one
An adaptive questionnaire that inferred the right healthcare package from a handful of answers, converting far better than the static form it replaced. Years before the vocabulary existed, this was already a recommendation problem.
2014 — 2016 · Thomson Reuters
data engineering, before the name
Crawlers and spider scanners processing legal documents at volume and high accuracy. Ankit didn’t call it data engineering at the time. It was.
how the toolkit changed
Where the weight of the work sat at each stop — and how the centre of gravity drifted from pages, to platforms, to models.
Relative emphasis, not a scorecard — a picture of where the days actually went.
what’s in the toolbox
The craft — and the people work that surrounds it.
Backed by a Master of Data Science, proven in production.
The parts that never belonged to only one track.
education
2024
Master of Data Science
Deakin University · Melbourne, Australia
2013
B.Tech, Metallurgical & Materials Engineering
National Institute of Technology, Trichy
beyond the résumé
Long before any of this, I was the kid who liked the problem more than the marks.
get in touch
If you have a hard problem on either side of that fork — a platform to build or a model to put into production — I’d like to hear about it.