shubham.richhariya
Staff Backend Engineer · Engineering Leader · 12+ yrs

Shubham
Richhariya

I turn messy, ambiguous problems into simple systems that ship, whatever the stack.

◇ Bengaluru, India ◇ Ex-Collective Artists · Redblock · Akamai · Oracle ◇ B.Tech CSE, NIT Bhopal

Currently building Weft, which weaves your social bookmarks into a searchable, auto-tagged library.

Get in touch View work ↓
$25M+
transactions
processed
High
availability
in production,
at scale
5TB+
lakehouse
/ 1.5M+ accounts
~10
cross-functional
team led
01

Profile

A backend engineer at heart, with 12+ years on the systems where mistakes are expensive: payments, secure APIs, and data platforms at scale.

That has meant a $25M payments pipeline and a multi-terabyte lakehouse, built from zero at startups, plus earlier backend work at Akamai and secure financial APIs at Oracle. More recently I went deep on agentic AI at Redblock, building multi-agent systems and computer-use agents (web navigation plus vision) that automate real work. I lead cross-functional teams and stay hands-on, still designing the hardest APIs alongside the team.

FocusBackend · APIs · Data platforms
Open toAI Eng · Backend · Leadership
Team led~10, cross-functional
CloudAWS (deep)
Patent-pendingLLM security work
Based inBengaluru, IN
02

Experience

Collective Artists Network · Head of Engineering
Sep 2024 – Apr 2026
Data Platform Infrastructure · Scale & Analytics Automation · hands-on
  • Made cross-platform social analytics possible: designed a 5+ TB lakehouse over 1.5M+ social-account records (follower counts, engagement, and profile data) on Dagster, Apache Iceberg, and StarRocks, with an analytics API answering core queries at sub-second latency and LLM workflows that auto-classified and enriched incoming data, replacing slow, ad-hoc reporting.
  • Halved the platform's AWS storage-and-network bill (50%) through deliberate cost engineering: Iceberg binpack compaction, snapshot cleanup, and a consolidated single-AZ layout on a self-managed EC2 NAT that removed cross-AZ and NAT-Gateway fees, a conscious cost/availability tradeoff for a batch analytics platform.
  • Scaled the engineering org's throughput: rolled out Jenkins CI/CD across core projects, migrated legacy JavaScript → TypeScript to enforce standards, and built a custom Jira hiring pipeline to grow the team.
  • Hardened access as the org grew: restructured AWS with Organizations & IAM Identity Center for centralized, auditable control, and cut licensing spend by standardizing on open-source.
Quolum → Redblock.ai · Founding Engineer, then Technical Lead
Aug 2020 – Sep 2024
One company through a pivot: fintech spend management, then agentic AI for identity security
Quolum: spend management & core payment infrastructure
  • Owned the revenue-critical path: engineered the end-to-end payments pipeline (Kinesis, SQS, SNS, Elasticsearch, Redis) that processed $25M in real-time purchases and refunds with near-continuous availability.
  • Took a fintech spend-management platform from zero to production as founding engineer: the core system and a company-card program built on Marqeta, with infrastructure in per-service CloudFormation stacks.
  • Cut failure rates 70% across 100+ third-party integrations by moving the long-running jobs off Lambda, where they were hitting timeout and concurrency limits, onto EC2, ending the on-call pages they caused.
  • Cut new SaaS integration delivery from 6 days to 2 days by building a custom code-generation toolchain: reusable Maven archetypes and Apache Velocity templates, adopted by 6+ developers to scaffold every integration, years before LLM codegen existed.
Redblock.ai: agentic AI for identity security, after the pivot
  • Cut customer security-remediation turnaround by 80% (weeks to hours): built AI agents that operate disconnected, API-less apps through their own UI (web navigation plus vision-model screen parsing) to detect and automatically remediate identity and access risks that traditional tools cannot reach, part of patent-pending work.
  • Architected the system behind it: a high-availability multi-agent orchestration framework with a central coordinator for task delegation, web navigation, and screen parsing; benchmarked OpenAI vs. open-source vision models for latency, accuracy, and cost.
  • Built the ingestion and transformation pipeline (DynamoDB, Kinesis) that aggregated users, roles, and access from across disconnected apps into the signals the agents acted on.
Embibe · Software Engineer II
May 2019 – Aug 2020
Content Management System (CMS) Infrastructure
  • Built the ingestion pipeline behind Embibe's CMS, organizing its educational-content catalog for review and publishing.
  • Shipped the content-approval workflow: quality checks and structured review that gated publication.
Akamai Technologies · Software Engineer II
Aug 2015 – May 2019
Momentum Order Management System (OMS)
  • Ran the Momentum OMS that Akamai's global Sales and Marketing teams depended on: order-management tooling built and maintained across Salesforce Classic, Lightning, and Mobile.
  • Led the Ruby on Rails upgrade and refactor of the legacy distributed codebase, retiring the technical debt that was slowing every new feature.
  • Cut server-side document generation time 60%: built custom async PDF/DOC/Excel generation for MSAs and sales documents.
Oracle Financial Services · Associate Application Developer
Aug 2014 – Aug 2015
Liquidity Management Infrastructure Product
  • Built the secure REST APIs behind real-time data integration for Oracle's financial liquidity product.
03

Projects

Weft

local-first open source

The problem: you save things across Instagram, X, LinkedIn, Reddit, YouTube, and more, and it all piles up siloed and unsearchable. A café you meant to try, a system-design thread, an article to revisit, then months later you cannot find any of it and just search from scratch.

Weft weaves your saves from 8 sources into a searchable, auto-tagged library, so you can actually find things again: search, ask questions over it (RAG), and draft from it. Runs 100% local; nothing ever leaves your machine.

PythonFastAPI PostgreSQL + pgvectorRedis Local LLMs via OllamaRAG React / TypeScriptChrome extension MCP server
04

Stack

Languages

JavaPython Spring BootTypeScriptJavaScriptReactRuby on Rails

Backend & APIs

REST APIsGraphQLFastAPI MicroservicesEvent-driven (SQS / SNS / Kinesis) RailsAPI design

Data Engineering

DagsterApache IcebergStarRocks TrinodbtDLT ParquetAthenaAWS Glue

Cloud · AWS

EKSS3RDSLambda SQSKinesisDynamoDB EventBridgeCloudFrontIAM Identity Center

Databases

PostgreSQLElasticsearchRedis DynamoDBpgvectorPinecone

Agentic AI & LLMs

Agentic AIMulti-agentComputer-use agents LLMsRAGVision Models LangChainOllamaMCP

DevOps & Infra

KubernetesKarpenterHelm DockerJenkinsCloudFormation ServerlessGit
Open to conversations

Let's build something resilient.

Open to agentic AI, backend, and engineering leadership roles. Whether you need AI agents that do real work, someone to own the hardest part of the system, or a lead for the team that ships it, I'd love to hear about it.

Email me LinkedIn ↗ GitHub ↗