Shivam Shahi

Software Engineer

New Delhi, India · Remote with a London-based team

Product-minded Software Engineer focused on building scalable, high-performance applications and robust system architectures.

2 mergedPRs in Aden Hive, a YC-backed 11k+ star repo

About Me

I'm a software engineer who likes owning a problem from the first sketch to production. Since 2023 I've worked across the stack, from the interfaces people use every day to the backend systems and infrastructure underneath them.

Most recently I joined Farben as the first engineering hire and helped take the product from early foundations to production. That meant the decisions that never show up in a demo: running isolated regions with proper data residency, bridging the cloud to on-premise accounting systems with a small Go agent, and building real-time pipelines for invoices and reconciliation. I do this remotely with a UK team, so async communication and time zone overlap are second nature. Before that I built AI agents, RAG and document pipelines at CognoTools.

I do my best work in small, fast-moving teams where one person can ship a feature, debug it in production and fix the root cause. I also contribute to open source, including merged PRs in Aden Hive, and write about system design. Away from the keyboard you'll find me playing soccer or exploring new music on Spotify.

Experience

A brief about my work experiences and professional journey.

•As the first engineering hire, I drove the technical vision and foundational architecture from inception to production.
•Shipped production as four isolated regions on GCP (UK, EU, India, Singapore) instead of one app with replicas. Each organisation lives in one region, with its own database, storage, keys and AI calls.
•Built data residency into the services: private-network Postgres with encrypted connections, per-tenant encryption keys managed in-region, and services that reject data belonging to another region. Held every region to the same strict compliance bar.
•Designed region routing so login finds the owning region through a single consolidated lookup (email to region) instead of fanning out to every region, with credentials only ever entered on the owning region.
•Architected a hybrid-cloud platform using a custom Go-based edge agent, seamlessly integrating cloud infrastructure with legacy on-premise ERPs like Tally.
•Engineered and automated end-to-end workflows for contract execution and document state synchronization via Slack and DocuSign APIs.
•Designed and implemented robust real-time SSE infrastructure, ensuring instant state synchronization for invoices, payments, and reconciliations.
•Developed scalable multi-stage ingestion pipelines for invoice processing, automated approvals, and complex journal entry reconciliations.

GCPMulti-regionData ResidencyEncryption & KMSPostgreSQLFastAPIRedisReactTypeScriptGo

Farben multi-region production: four isolated regions, not one app with replicas

I shipped Farben, a multi-tenant finance product, as four isolated GCP regions with real data residency. One organisation lives in exactly one region. Login finds the region; signup does not. Hover or tap the drawing to see what each part does.

REGION · EU · NETHERLANDS

A complete, isolated copy of the stack in the Netherlands: its own compute, database, storage, keys and AI calls. An organisation created here lives here, and nothing replicates out.

Region: europe
Isolation: Own stack, own data, own keys
Stays: Data, files, keys and AI calls in-region

DRAWN BY	Shivam Shahi
PROJECT	farben.ai
REV	2026.11
SHEET	1 of 1

Residency

Each region owns its database, storage, keys and model calls. No shared application data and no replication, and the service refuses data that belongs to another region. The only shared piece is a minimal routing directory.

Encryption

TLS on every host, a private-network database that only accepts encrypted connections, and per-tenant keys managed in the region itself.

Routing

One organisation, one region. Login finds the region, signup does not. A small routing directory maps an email to its region in one lookup, and the password is entered once, on the owning region.

Compliance

Each region's residency and data-handling requirements were hard constraints, not best effort: enforced in code, encrypted in transit and per tenant, access audit-logged, and held to the same bar in all four regions.

Read the full write-up

Open Source

Merged pull requests in production-grade open-source projects.

aden-hive/hiveYC-backed

Multi-Agent Harness for Production AI

11k+ stars

0
PRs merged: +0
Lines added: 0
Files changed: 0
Integrations shipped

Also merged: 100xdevs CMS #69 (new “Daily Code” section). More on GitHub.

Projects

Some of the projects I've built and deployed.

ShadowLLM

Production AI reverse proxy for zero-latency traffic bifurcation. Safely shadow-tests experimental prompts and cheaper models against live production traffic with side-effect virtualization and real-time semantic drift scoring.

TypeScriptFastifyRedis StreamsPostgreSQL (Neon)+8

Unified Multi-Cloud Python SDK

Multi-cloud serverless developer SDK and CLI providing a single unified API abstraction over disparate serverless runtimes (Cloudflare Workers, AWS Lambda). Features AST code serialization, automated packaging, local emulation runtime with FastAPI, and extensible provider adapter architecture.

Python 3.11SDK ArchitectureAdapter PatternAST Code Serialization+5

Payment Processing Engine

High-throughput concurrent payment orchestration engine in Java 17 LTS. Features atomic multi-key idempotency locks, fine-grained thread synchronization, dynamic multi-currency fee calculation strategies, state machine transition guards, and resilient exponential backoff retry policies.

Java 17Low-Level Design (LLD)SOLID PrinciplesConcurrency & Thread Pools+6

Concurrent Task Scheduling Engine

Thread-safe asynchronous task execution engine with Directed Acyclic Graph (DAG) cycle detection dead-lock prevention. Implements deterministic priority queues with atomic sequence tie-breaking, bounded queue admission control, worker thread fault isolation, and graceful zero-leak shutdown.

Java 17MultithreadingDAG Cycle Detection (DFS/BFS)Priority Blocking Queues+5

Realtime Collaborative Kanban Board

A Kanban Board Application built with React, Redis, and WebSockets for real-time task management and multi-client board synchronization.

ReactJavaScriptSupabase (Postgres)Upstash (Redis)+6

Ai Symphony

A platform providing AI tools like image generation, summarization, and chat using open-source models deployed on edge workers.

TypeScriptTailwind CSSNext.jsCloudflare Workers+4

Bug Issue Tracker

A comprehensive bug tracking system with role-based authentication, project management, and real-time issue tracking.

ReactNode.jsExpress.jsSQLite+5

Writing

All articles

Deep dives into distributed systems, backend architectures, and real-world system design.

HashnodeSeptember 29, 202616 min read

How YouTube Processes a Video After You Upload It

From a 2 GB upload to transcoding, queues, adaptive bitrate streaming, and CDN delivery.

“You open YouTube Studio, pick a 2 GB file, and click Upload. The bar fills. A few minutes later the video plays in more than one quality, on a phone, a laptop, and a television. That looks like one operation. It is not. Between the click and the first frame someone else can stream, the platform has to accept a large file over a bad network, store it, encode several versions, build metadata, survive machine failures, and serve the result from an edge CDN near the viewer.”

System DesignDistributed QueuesAdaptive Bitrate (HLS/DASH)TranscodingEdge CDN

Technical Arsenal

I view languages and frameworks simply as tools to solve complex business problems. While I adapt quickly to any ecosystem, I have deep production expertise in a few core areas:

Frontend & Product: React, Next.js, TypeScript, and modern state management. I build interfaces that feel instantaneous and native.
Backend & Systems: Node.js, Python, and Go. I design scalable microservices, real-time event-driven architectures, and secure API gateways.
Data & Infra: PostgreSQL, Redis, Docker, and hybrid-cloud deployments. I focus on high availability, efficient querying, and resilient deployments.