software engineer · full-stack builder
I ship production backend and platform systems for enterprise SaaS — multi-cloud infrastructure, identity, and security that run across AWS, GCP, and Azure in 10+ regions. Off the clock, I build whatever a problem needs: AI-agent infrastructure, Android and iOS apps, CLIs, and ML pipelines.
scroll to traverse to the next checkpoint ▾
Five years across enterprise security, fintech, and platform infrastructure — owning services end to end.
I owned multiple core backend services (Go & Python) for a multi-tenant, browser-native enterprise security platform on GCP, AWS, and Azure. A few of the systems I built and led:
A node-orchestration control plane that provisions, autoscales, and upgrades compute across AWS, GCP & Azure in 10+ regions — with a warm instance pool and spot-eviction handling to cut cloud spend.
Built largely solo: ingests users, groups & devices from five identity providers — Entra, Okta, Google, JumpCloud, Ping — into per-tenant Postgres + BigQuery over gRPC.
A multi-tenant OpenID Connect / OAuth2 provider — full authorization, token & JWKS surface with per-tenant isolation and federated login via Entra & Ping.
A six-service Go suite for credential management, domain categorization, WHOIS, and breach detection on GKE — fed by a Kafka / Pub-Sub event multiplexer.
Plus the policy & access-control layer (policy engine, DLP, SSO, members/groups/devices), the deployment toolkit that stands up per-tenant infrastructure across five regions, and the reliability layer underneath it all — CI/CD, observability, SLO alerting, and secret management.
end of intel — keep scrolling to traverse to the next checkpoint ▾
Real, shipped side projects, built for the love of it — from AI-agent infrastructure to mobile apps, CLIs, and ML pipelines.
Live products
A storefront that turns impulse buying into saving. Browse and "buy" like any e-commerce PWA — but checkout deposits the money into your own investment account instead of shipping goods. You pay yourself.
A voice-driven agent for the Mac. Speak to it and it acts on your machine — transcribes your voice locally with an LLM, routes the request to a frontier model that can run real commands, and talks back with system speech, with a leash that holds risky actions for confirmation.
AI & agent infrastructure
Agent-first version control — a greenfield VCS (no git underneath) built for autonomous coding agents. Encrypted visibility zones keep private code private by construction, credentials are sealed objects an agent can use but never read, workspaces fork in O(1) for cheap parallel agents, and merging never fails — concurrent edits produce a legal merged state with a first-class conflict object. A blind-by-default server stores only opaque blobs and hydrates content on demand. Driven over MCP or a CLI.
A structured personal-knowledge service that any AI agent can query over MCP (Model Context Protocol). It stores preferences, policies, vendor ratings, and order history across domains — food, coding, cabs, services — so an agent calls get_briefing() for fast, token-frugal context before it acts. Built with Go, Postgres + pgvector, and MCP over HTTP/SSE, with a React dashboard for feedback and metrics. Includes sensitivity classification, provenance tracking, and idempotent markdown import.
Live Hindi sales-call coaching. Two people role-play a loan call with push-to-talk; it transcribes via local whisper-cpp (Metal-accelerated), whispers real-time nudges, then scores the call across 7 dimensions with evidence-verified quotes. Runs entirely locally — no API keys, no cloud.
A local AI-agent manager — like nvm, but for open-source AI agents. Install, configure, and launch tools like Ollama and OpenCode. No API keys, no cloud — everything runs locally. Named after the sage Lomasa from the Mahabharata.
Apps, tools & ML
Turns event photos into a deployed site where guests click a face to see every photo that person appears in. RetinaFace + ArcFace for detection and recognition, agglomerative clustering to group faces, collaborative naming. Deployed on Cloudflare Pages + R2 + KV with password protection. Used at real weddings and events.
Bilingual (English + Hindi) Android app and FastAPI backend giving farmers in Kanpur Dehat, UP actionable insights — IMD weather, severe-weather nowcasts, and advisory bulletins parsed from DAAS PDFs. Every card shows its source and freshness. Offline-first with WorkManager sync. In internal testing.
A browser extension that adds a "skin" to YouTube — save videos and channels without logging in, solving the problem of losing recommendations when watch history is off. Published on both the Firefox Add-ons and Chrome Web Store.
A rental marketplace where owners and renters connect directly. Full-stack: a Kotlin Android app with a FastAPI backend, containerized with Docker. A complete product, pitch deck and all.
end of intel — keep scrolling to traverse to the next checkpoint ▾
I don't stick to one platform. Android in Kotlin, iOS in Swift, browser extensions in JS, CLIs and backends in Go, services and ML pipelines in Python. If it solves a real problem, I'll learn whatever stack gets it done — and ship it.
My day job is backend and platform engineering at scale: multi-cloud orchestration, identity and access, and the reliability work that keeps multi-tenant systems running across regions. The throughline is ownership — I like holding a service end to end, from the data model to the deploy pipeline to the 2am alert.
Right now I'm deep in the AI/LLM space — not just using these tools, but building infrastructure around them. second-brain is an MCP-native memory service for agents; sales-saathi runs local whisper + LLM for real-time Hindi coaching; lomasa-ai is where the open-source work lives. I build for domains I care about — agritech, AI tooling, and the small tools that make daily life better. Human ideas, AI-assisted execution.
Enterprise security · Identity & access · Multi-cloud orchestration · Fintech · Agritech · AI/LLM tooling
GCP · AWS · Azure · Kubernetes/GKE · Docker · Terraform · gRPC · Kafka/Pub-Sub · Postgres · BigQuery · Elasticsearch · Redis · Cloudflare
Tvash is my AI collaborator — a Claude-powered agent that's an actual part of my workflow, not a gimmick. The lomasa-ai projects are human–AI pair-programmed: I bring the ideas and judgment, Tvash helps execute. @claudeforssb
end of intel — keep scrolling to traverse to the next checkpoint ▾
// final checkpoint — channels open
© Shivam Som Bajpai
end of patrol — or scroll up to retrace the path ▴