● akshay.nanavare

menu

SELECT * FROM projects

Projects

Case studies from production work: the problem, what I built, and what changed.

★ featured

Last9 · Zero-downtime ClickHouse migration

50+ live tenants migrated with zero downtime and significantly lower compute cost.

problem: Compute cost grew with every tenant we added, and customers rely on ingestion, search and alerts around the clock, so downtime was not an option.
approach: I designed the migration strategy, mentored by our CTO, and three engineers joined for the rollout.

ClickHouse
Altinity
Kafka
Kubernetes

Last9 · APM backend

Became one of the product's core features; the MVP contributed to product-market fit.

problem: Customers wanted application performance monitoring next to their logs and traces, in one product.
approach: Designed and built the APM backend from scratch as an MVP, on top of the traces pipeline and ClickHouse storage, then grew it into a core feature.

Go
ClickHouse

Last9 · Ingestion control plane and physical indexes

Customers trim data before paying to store it, and indexed queries scan only the data they need.

problem: Customers paid to store data they didn't need, and a filtered query (say, only staging logs) still scanned everything.
approach: Ingestion-time rules (extract, remap, drop) that run before data is written, plus user-defined physical indexes that route matching data into its own dedicated table end to end.

Go
ClickHouse
OpenTelemetry

Last9 · Cold storage and rehydration

Long retention at object-storage prices, with archived logs queryable again whenever they're needed.

problem: Customers need months of logs for audits and incidents, but keeping it all hot in ClickHouse is expensive.
approach: Back up logs to the customer's own object storage (S3, GCS), then rehydrate the time range they ask for back into the product on demand.

Go
ClickHouse
AWS S3
GCS

Last9 · Tenant onboarding automation

New tenants, SaaS or BYOC, come up without manual infra work.

problem: Every new customer needed infrastructure brought up by hand, for both SaaS and BYOC (bring your own cloud) setups.
approach: A SaaS signup or BYOC setup triggers automated provisioning of the tenant's infrastructure on Kubernetes.

Go
Kubernetes

Last9 · Event-driven S3 log ingestion

Logs from S3 become queryable in near real time.

problem: Customers already ship logs to S3 and wanted them searchable without running another agent.
approach: An event-driven pipeline that reacts to new objects in S3, parses them, and ingests them into the logs store.

Go
AWS S3
ClickHouse

Last9 · ClickHouse SQL query generator

Powers log querying in the product, with query performance tuned for large-scale logs and traces.

problem: Users needed to filter, parse, and transform logs without writing ClickHouse SQL by hand.
approach: A Go query builder that compiles filters, parsers (regex, JSON, logfmt), and transformations into optimized ClickHouse SQL.

Go
ClickHouse

Last9 · MCP server for production debugging

Real-time debugging from Cursor, VS Code, and Claude.

problem: Engineers debugging production context-switch between their editor and the observability UI.
approach: An MCP server that exposes logs, traces, and exceptions to AI assistants.

Go
MCP

Tazapay · Transaction report pipeline (CDC)

Report generation went from 15 minutes to under 1 minute.

problem: Merchant transaction reports took 15 minutes to generate.
approach: A new Go service fed by change data capture over AWS Kinesis, so report data is prepared as transactions happen.

Go
AWS Kinesis
CDC
gRPC

Tazapay · Ops control panel

80% fewer ops tickets.

problem: Every change to a merchant's payment methods needed an ops ticket to engineering.
approach: A Go control panel that lets the ops team manage merchant payment methods themselves.

Go
gRPC