Forward Deployed Engineer · Rill Data
I take enterprise customers from first call to production analytics, and I engineer every layer in between, from the data model to the Kubernetes cluster.
I run 2+ PB on Kubernetes across ClickHouse, Apache Druid, and DuckDB, and I’ve taken 50+ enterprise customers live after 150+ proofs of concept, owning everything from SSO and migrations to storage layouts, query tuning, Kubernetes operators, and customer success. One of Rill Data’s earliest engineers.
SELECT metric, value FROM rohith.profile;
| metric | value | note |
|---|---|---|
| years_in_data | 9+ years | since 2016, across four companies |
| years_forward_deployed | 5+ years | at Rill Data, since 2021 |
| enterprise_customers | 50+ | from scoping to production and support |
| customer_pocs | 150+ | run with Sales and GTM |
| data_managed | 2+ PB | ClickHouse, Druid, and DuckDB on GKE |
| olap_clusters | 30+ | 17 Druid and 14 ClickHouse, sharded and replicated |
| storage_cpu_saved | 30–40% | from schema, compaction, and right-sizing |
| tb_scan_time | <30s | TB-scale scans after sort keys, projections, and codecs |
8 rows in set. Elapsed: 0.004 sec. Processed 2.00 PB.
Public Rill customers, via rilldata.com
01 · What I do
Forward-deployed work means owning the whole technical problem, from how a query reads parts off disk to how the cluster is scheduled on Kubernetes. These are the areas I own in production.
Scoping to go-live for 50+ enterprise customers, after 150+ POCs.
Each user sees only the data they should.
Moving data in both directions, with no data loss.
Production analytics keep running during upgrades.
~30–40% less storage and CPU, and TB-scale scans in under 30 seconds.
Batch and real-time ingestion at GBs per hour.
Tooling that makes each deployment faster than the last.
2+ PB across ClickHouse, Druid, and DuckDB.
02 · Product & customer success
Forward-deployed engineering is half engineering and half product and customer work. I prove the product works, bring what customers need back into the roadmap, and stay the technical owner of each account after go-live.
150+
POCs with customers
I prove the product works on a prospect’s real problem before they commit.
50+
accounts feeding the roadmap
I bring what customers need back into the product, and help ship it.
50+
enterprise customers onboarded
I stay the technical owner of the account long after go-live.
03 · How I operate
What I’m like to work with, and the evidence behind each claim.
I joined Rill Data in 2021 as one of its earliest engineers. Nobody had written the playbook for running petabyte-scale real-time analytics for enterprise customers, so I built the infrastructure, the processes, and the playbook as we went.
Evidence
At Rill I sit between the customer, Sales, Product, and Engineering. Rill’s customers include The Trade Desk, Comcast, AT&T, AppLovin, and FreeWheel: demanding AdTech and media teams whose dashboards drive revenue every day.
Evidence
When a gap costs the team time or money, I don’t wait for a ticket. I measure it, fix it, and make the fix stick with process or tooling.
Evidence
Most of my work starts with an unfamiliar customer stack and an unclear problem. I get hands-on fast, build a working version, and let the structure emerge from what I learn.
Evidence
A fix isn’t done until the next person doesn’t need me. Every change I ship comes with the runbook, README, or guide that explains it.
Evidence
Outside my day job I build, publish, and keep improving things, from a tutorial site for developers to hackathon products and open-source tools.
Evidence
04 · How I work
Every one of my 50+ enterprise customers, and the 150+ POCs before them, followed the same shape. I own each step, and the customer always knows where we are.
A proof of concept on the customer’s real problem, with success criteria agreed up front.
Requirements with technical and executive stakeholders, turned into a data architecture.
SSO, row-level security, and least-privilege access to cloud storage.
Connect or move data from legacy systems, warehouses, and lakehouses.
Batch and real-time ingestion, models in ClickHouse or Druid, and dashboards.
GitOps-managed dedicated cluster or a BYOC install, taken live.
Train users, hand off to steady state, and stay the technical owner of the account.
05 · Experience
From ERP migrations to petabyte-scale real-time analytics. Every role added a layer: migrations, then distributed compute, then streaming, then running it all for enterprise customers.
Senior Data Engineer, Forward Deployed & Solutions Engineering
One of the earliest engineers at Rill, an open-source, real-time operational BI platform. I own forward-deployed engineering and petabyte-scale data infrastructure for enterprise AdTech and marketing-analytics customers, and I owned US time-zone engineering coverage on a distributed team.
– Present
5 yrs 5 mos · Boston, MA · Remote (US)
Data Engineer
Streaming, warehousing, and ML for a classifieds marketplace.
–
2 yrs · Bengaluru, India
Software Development Engineer
Data-intensive analytics over industrial IoT sensor data.
–
1 yr 3 mos · Bengaluru, India
Technical Associate
Where my data-migration work started.
–
1 yr 2 mos · Hyderabad, India
06 · Projects
Case studies from petabyte-scale work for enterprise AdTech and marketing-analytics customers. Each one covers the problem, what I built, the impact, and the skills behind it.
A self-hosted, multi-tenant ClickHouse platform on Google Kubernetes Engine that runs 14 production ClickHouse clusters for enterprise real-time analytics, as part of a 2+ PB OLAP fleet.
An MCP-based AI toolkit that lets Claude Code audit and optimize ClickHouse clusters across the fleet, finding the columns, tables, and queries that drive storage and compute cost and tracing each one back to the data model that created it.
Led migrations of enterprise customers from a legacy analytics platform and older pipelines to the new Rill application, moving data, metrics, and dashboards without disrupting daily reporting.
Streaming and batch pipelines that ingest high-volume programmatic-advertising data (bid requests, wins, impressions, and auctions) into Apache Druid and ClickHouse for real-time analytics.
Migration paths that move analytics data from Snowflake and lakehouse tables into ClickHouse and Apache Druid, and back out through reverse ETL, with automated checks that confirm the numbers match at every hop.
Re-architected a high-volume ClickHouse analytics workload around tiered projections, directory-based partitions, and tuned compression, which stopped out-of-memory dashboard failures and cut storage and compute.
A production AI on-call agent that turns monitoring alerts into triaged, deduplicated tickets and Slack updates, so engineers start from a diagnosis instead of a raw alert.
The observability layer for a 30+ cluster, 2+ PB ClickHouse and Druid fleet: templated monitoring dashboards, data-freshness alerting, SLOs, and a usage-billing pipeline that shows customers exactly what they consume.
Public code: the platform I help build, plus tools and examples I’ve published.
Open-source, real-time operational BI on ClickHouse, Druid, and DuckDB.
Agent skills that let Claude Code and Cursor develop, query, and debug Rill projects.
Idempotent incremental ingestion into ClickHouse with Rill, as a runnable example.
Programmatic-advertising bid logs analyzed with the OpenRTB framework in Rill.
A CronJob Kubernetes operator in Go, built with Kubebuilder, with a step-by-step guide.
Automatic discovery of Kubernetes services as Prometheus scrape targets.
07 · Stack & credentials
Highlighted tools are the ones I run in production every day. The rest I’ve shipped with and can pick back up quickly.
M.S., Data Architecture & Management
Northeastern University · 2023 – 2025
Executive MBA, Digital Marketing & Analytics
Indian School of Business
B.Tech, Electronics & Communication Engineering
Amity University · 2012 – 2016
Blog
Deep dives on problems I’ve solved in the field.
20 min read
A column that is mostly small with the occasional huge value. Should it be Int64 with a T64 codec, a Variant(Int8, Int64), or something else? How the major OLAP engines handle integer width on disk and in memory, with measurements from ClickHouse and DuckDB.
11 min read
How I converted BigQuery’s ZetaSketch HLL++ sketches into ClickHouse uniqCombined64 states by transplanting the registers directly: no rehashing, no raw rows, and counts within HLL’s own noise.
Quick answers
Short, direct answers about what I do and how I work.
All questionsRohith Reddy Kota is a Boston-based Forward Deployed Engineer and Senior Data Engineer at Rill Data, and one of its earliest engineers. He manages 2+ petabytes of data on Kubernetes across ClickHouse, Apache Druid, and DuckDB, and deploys, migrates, upgrades, and supports real-time and batch analytics for enterprise AdTech and marketing-analytics companies.
He leads enterprise customer deployments end to end. That covers scoping, SSO and security, data migrations, onboarding and training users, GitOps-managed dedicated Kubernetes clusters, system and database upgrades, production support, and handoff. He has deployed 50+ enterprise customers, run 150+ proofs of concept, and runs 30+ ClickHouse and Druid clusters on GKE.
He scopes requirements with technical and executive stakeholders, designs the data architecture, sets up SSO and cloud-storage access, and migrates or connects the customer’s data. He then builds batch and real-time ingestion, models the data in ClickHouse or Druid, builds dashboards and alerts, provisions a GitOps-managed dedicated cluster on Kubernetes, trains the customer’s users, and hands off to steady-state support.
He has deployed and supported 50+ enterprise customers on Rill and run 150+ proofs of concept with prospective and existing customers, mostly in AdTech and marketing analytics. Public Rill customers include The Trade Desk, Comcast, AT&T, AppLovin, FreeWheel, InMobi, Moloco, and Liftoff.
He manages 2+ petabytes of data on Kubernetes across ClickHouse, Apache Druid, and DuckDB, including 30+ distributed OLAP clusters (17 Apache Druid and 14 ClickHouse) on Google Kubernetes Engine. The work covers constant batch and real-time ingestion, ingestion-failure recovery, upgrades, migrations, and observability.
He treats cost as a first-class production metric. He right-sizes clusters and node pools, cuts storage through compression, retention, TTLs, and compaction, removes unused data, and makes queries cheaper. He also built a usage-billing pipeline for visibility. These changes cut storage and CPU by ~30–40% for enterprise customers.
He runs a production AI on-call triage service: a FastAPI application that takes monitoring webhooks, classifies issues with an LLM using structured outputs, and opens or updates tickets and Slack threads. He has also built MCP servers and agent skills that let Claude Code and Cursor develop and query Rill projects.
His major projects include a self-hosted ClickHouse platform on Kubernetes running 14 production clusters and an MCP-based AI toolkit that audits and optimizes ClickHouse storage and cost. He has migrated enterprise customers from a legacy analytics platform to Rill, built real-time AdTech bidstream pipelines on Kafka, Apache Beam, and Dataflow, and moved Snowflake and lakehouse data into ClickHouse with reverse ETL and integrity checks. He also redesigned a ClickHouse workload around projections, built a production AI on-call triage agent, and built fleet observability with data-freshness alerting and usage billing.
08 · Contact
If something here resonates, whether it’s a data problem, an idea, or a conversation worth having, I’d love to hear from you. The fastest way to reach me is a message on LinkedIn.
Good reasons to reach out
Data platform help
ClickHouse or Druid performance, Kubernetes operations, migrations, and cost reviews.
Collaboration
Open-source work, writing, talks, or comparing notes on a hard data problem.
Just to talk shop
Real-time analytics, AdTech data, MCP and agent tooling, or life as an FDE.
