TURTLEBYTE — BENGALURU, INDIA
A small engineering team in Bengaluru, working with clients in the US, UK, and Australia. We operate what we ship, not just build it.
Our own dating app, Amour, is live on Android — Google Play↗
React Native · Next.js · Node · Python · PostgreSQL · Kubernetes · AWS
01 — CASE STUDY
Our own product
Amour is our product, not a client's. We designed it, built it, and run it in production, which means every decision below was ours to make and ours to live with. It is live on Android.
A consumer dating product for the Indian market needed a mobile app and a backend that could handle real-time messaging, heavy media upload, automated content moderation, subscription billing, and identity verification — with a small team and a tight infrastructure budget.
Microservices communicating over gRPC. PostgreSQL, accessed through Prisma, is the primary store; Neo4j holds the matching and relationship graph; DragonflyDB handles caching; RabbitMQ carries asynchronous work between services.
Chat and activity feeds run on GetStream, push notifications on Firebase. Media is stored on Cloudflare R2 behind an upload and processing pipeline.
Image moderation through AWS Rekognition and text moderation through AWS Comprehend, both running asynchronously off the message queue so nothing blocks a send.
Identity and access are handled through Logto. Subscriptions and entitlements run through RevenueCat.
The Android app is React Native. Redux Toolkit and Redux Saga carry state and secure session persistence, gRPC-Web gives typed communication with the services above, and Agora handles real-time messaging with offline queuing.
Self-hosted on K3s running on Hetzner, with deployments managed declaratively through Flux GitOps. We chose this over managed cloud to keep hosting costs an order of magnitude lower without giving up reproducible deploys.
STACK
MOBILE
DATA
MESSAGING
MEDIA & SAFETY
INFRA
Shipping a consumer mobile app together with the services under it: service decomposition, async pipelines, third-party integration at depth, and owning infrastructure end to end rather than handing it off.
02 — CASE STUDY
Contract engagement, Avacend Labs
A healthcare facilities analytics product. We were brought in under contract to build the data and analytics layer end to end, from ingestion through to the dashboards operators actually use.
Facilities teams sat on years of work-order history locked inside a third-party system, reachable only through its API. Answering an operational question meant exporting spreadsheets by hand. They needed that data continuously synced, queryable, and readable by people who do not write SQL.
Python pipelines pulling from the WorxHub API over OAuth2, with automatic token refresh and a three-tier extraction strategy — full cumulative loads, incremental deltas, and on-demand pulls — so routine syncs stay cheap and backfills stay possible.
A ClickHouse schema built on ReplacingMergeTree so repeated extractions converge instead of duplicating. Loading runs as batched inserts across parallel workers with exponential backoff, which keeps large syncs from stalling on transient failures.
A text-to-SQL pipeline over GPT-4o translates a plain-language question into ClickHouse SQL, runs it, and interprets the result back into readable prose. Operators ask for what they want rather than learning the schema.
Work orders are embedded with SentenceTransformers, reduced with UMAP, and clustered with HDBSCAN to surface recurring failure patterns nobody had labelled. Vectors are held in Pinecone for similarity lookups.
A multi-tenant FastAPI backend with tiered role-based access and JWT auth, serving four production dashboards covering SLA compliance and shift-based analytics. ClickHouse calls run off the event loop, and DragonflyDB caches results with invalidation keyed to extraction dates.
STACK
DATA
AI & ML
BACKEND
INTEGRATION
Owning a data platform from third-party API through to the interface: incremental sync that holds up, a warehouse schema that stays correct under repeated loads, and LLM and embedding work applied to a real operational problem rather than bolted on.
03 — WHO YOU'RE WORKING WITH
We keep the team small on purpose. The engineer who scopes your project is the one who builds it, and has spent four years shipping and operating production systems.
Tilak Kumar
Founder — TurtleByte
Founding engineer
First engineer on a distributed log management platform. Go microservices talking over gRPC, a StarRocks and Kafka ingestion pipeline handling high-volume log data, the Next.js front-end that visualised it in real time, and the Kubernetes setup that ran the whole thing.
Senior engineer, contract
The FM360 data and analytics platform above, built end to end under contract.
04 — SERVICES
React Native applications for iOS and Android, from first build through store release and the updates after it.
Next.js and React front-ends, internal dashboards, and the APIs they run on.
Designing and building services, APIs, and data models for products expecting growth.
Kubernetes, CI/CD, GitOps, and cost-efficient self-hosted setups.
Running, extending, and maintaining systems already in production.
Working as an embedded part of an existing engineering team.
05 — HOW WE WORK
We keep four hours of daily overlap with US and European working hours.
Most engagements begin with a fixed-price two-week piece of work, so you can judge the output before committing to anything larger.
For agencies, we work under your name and stay invisible to your client.
You talk to the engineers doing the work, not an account manager.
Thirty minutes, no deck. Bring the architecture problem you’re stuck on.
