Siranjeevi Subramaniyan

Open to full-stack, backend or platform roles

Full-stack and backend engineer, 3.5+ years Co-founder and Technical Lead, Estospaces

Co-founded Estospaces, a property platform, and led its build: 7 Go services on Google Cloud, a React web app and a Flutter Android app. Based in Salem, Tamil Nadu; open to Chennai, Bengaluru or remote.

IST, UTC+5:30

Siranjeevi Subramaniyan

01 Selected work

Estospaces

A property platform for India and the UK. Co-founder and Technical Lead, May 2025 to now.

  • 72/72

    production smoke checks after each of 11 releases, at launch-readiness in October 2026.

  • ~330

    REST endpoints

  • 6,000+

    automated tests across Go, React and Flutter

The Android build is on Google Play internal testing. The platform has not launched publicly, and production is offline as of November 2026. Every screen in this section comes from the dev environment with test data (the QA pass of 6 to 8 October 2026 unless a caption says otherwise); the Android screens below are the app's own test renders.

Looking for a full-stack, backend or platform engineering role. Full-time since May 2023: now Co-founder and Technical Lead at Estospaces, before that CourSea Academy and Enkindle Technologies.

I designed the architecture and led the build. Teammates built the first scaffolds, the first auth pass and the .co.uk landing site; I built most of the rest with AI coding agents that I directed and reviewed.

01b Estospaces · Android

The Android app

These six screens are rendered from the app's own code by its golden tests, at Play Store size, with a fictional demo cast. 74 golden images guard the interface.

Search results for two-bed rentals in Camden, with one verified listing
01 Search results.
Home detail for a two-bed flat near Camden Market, with its manager and a Book viewing button
02 Home detail.
Book a viewing: the calendar for October 2026 and the times still free on the chosen day
03 Book a viewing.
Fast Track at step 2 of 6: the manager asks for a newer proof of address
04 Fast Track: step 2 of 6.
Chat with the letting agent while offline: the last message was not sent and offers Tap to retry
05 Chat, with a message waiting to resend.
Manager app, Today: a new lead with 6 minutes left to reply and a viewing request
06 Manager app, Today.
Read the case study

Stack with production depth: shipped, tested, in git

02 Estospaces · system

The thirteen repositories

~830 merged pull requests across them between February and November 2026. Each bar is a real repository, grouped by layer; its length follows its commit count. The Edge and Data layers, listed in the table, are shared Google Cloud services.

Estospaces repositories by layer, with commit counts
RepositoryCommits
Clients
estospaces-webReact 19 and TypeScript web app1,354
estospaces-mobileFlutter Android app672
estospaces.comNext.js 16 marketing site and blog88
Edge
HTTPS load balancer, managed SSL, Cloud Armor WAF
Go services on Cloud Run + shared library
core-serviceIdentity, properties, leads, verification318
booking-serviceViewings, applications, Fast Track246
media-serviceCloud Storage uploads, signed URLs103
payment-serviceRazorpay subscriptions and webhooks94
notification-serviceIn-app notifications90
messaging-serviceConversations and support tickets90
search-serviceSearch and saved-search alerts72
sharedGo library, signed permits19
Data
One PostgreSQL 16 database per service on Cloud SQL, Cloud Storage, Secret Manager
Delivery
infrastructureTerraform, 11 modules149
deploymentGitHub Actions and Cloud Build, security scans, canary 5% to 100%46
Total3,341

Commits on all branches by my account and by the bot account my AI agents commit as, merge commits excluded (git rev-list, November 2026). An underlined name links to the place on this site that shows that repository's work. The Estospaces code is private, in the company's GitHub organisation; the case study prints excerpts.

03 Estospaces · process

How I direct AI agents

Claude Code and Codex run in separate git worktrees under a written RULES.md. Every change gets a risk class from R0 to R4. High-risk changes need an independent verifier agent, nothing merges without the checks and evidence its risk class requires, and releases go to 5% of traffic first.

The October 2026 launch review lists 17 high-risk (R3) pull requests in core-service and booking-service, each with an independent verifier pass. Two verifier rounds failed first, and those changes were fixed before release.

ClassExamplesRequired behavior
R0: Read-onlyExplanation, code reading, logs, plan reviewNo mutation; cite inspected evidence
R1: LowDocs, comments, copy, isolated style changeTargeted check; diff review; no independent verifier unless requested
R2: StandardUI behavior, ordinary endpoint, non-sensitive business logicRegression test, repo gates, relevant role/journey verification
R3: HighAuth, authorization, profile persistence, contracts, payments, uploads, messaging privacy, cross-service contract, dependency upgradeIndependent verifier, negative tests, security review, integration/E2E evidence
R4: CriticalProduction, Terraform apply, DB migration, secret rotation, region cutover, DNS/traffic switch, deletion, data backfillExplicit approval, written runbook, backup/rollback evidence, independent verifier, live monitoring

If uncertain, choose the higher class.

RULES.md, section 7, verbatim

Vocaryn

04 Vocaryn

An AI that finds matching jobs, tailors your application and submits it where you allow it. Live at vocaryn.tech, with paid plans since October 2026. Early stage.

A sent application inside Vocaryn: fit 90 of 100, the three answers that were entered, and the line that every sentence cites a verified Career Fact
A sent application: why the job fit and what was entered. From the test round of 1 November 2026: the production build run locally, with fictional accounts and employers.
Vocaryn agent home: I need you for 13 quick answers, and the question What is your notice period with four answers
Agent home: it asks the user instead of guessing an answer.
Vocaryn applications list on a phone: the Full Stack Engineer application at Orbit Analytics waits for the user's answer to what is your notice period
The same question, waiting in the app on a phone.

Truth Lock is a rule checker, not a second model. Every sentence the AI writes has to cite the user's verified Career Facts. A sentence with no citation, or with a name, a number or a technical term those facts do not contain, is rejected.

export type RejectionReason = 'no-citation' | 'unknown-fact' | 'unsupported-number' | 'unsupported-term' | 'too-long' | 'empty';
Vocaryn, packages/tailor/src/truth-lock.ts, line 5, at bb17e21 (17 October 2026), verbatim: the six reasons a sentence is rejected.
  1. I built the booking flow in React. [F4]cited: kept
  2. I would be a great fit for this team.no-citation
  3. I shipped the search service. [F99]unknown-fact
  4. I cut page load by 70%. [F4]unsupported-number
  5. I ran it on Kubernetes. [F4]unsupported-term
  6. A sentence that runs past the length limit…too-long
  7. (an empty sentence)empty
Illustration of the rule with made-up sentences, not app output. [F4] stands for a citation of one of the user's verified Career Facts.
  • Measured on 28 September 2026 in my own test: it caught 98.5% of fabricated items (707 of 718) and blocked 0 of 450 honest ones. It catches invented names and numbers, not vague wording, and the app says so on every application.
  • ~34,000 jobs from 479 company job boards
  • Auto-apply agent and Chrome extension for LinkedIn, Naukri, Indeed and 6 ATS sites. It never solves CAPTCHAs.
  • Self-hosted Supabase with row-level security, Cashfree payments, Cloudflare Pages
Visit vocaryn.tech

EstoGrowth

05 EstoGrowth

An AI marketing agent for small businesses, first called Markegent and then Loudgent. It is built so agents plan, write, design, publish, reply and report, and anything risky waits for a person.

EstoGrowth Outreach view for the Estospaces space: three outreach plays, one active and two proposed and paused
Outreach inside the app. Captured from a local copy with every outside connection switched off.
  • 136 commits in 9 days, November 2026: ~12,500 lines of TypeScript on Node 22 and SQLite, with 31 test files
  • One router across Claude, Codex, Gemini, Groq, Workers AI and Ollama
  • Publishers for 9 social platforms through their official APIs. Accounts for LinkedIn, Bluesky and YouTube were linked on 15 November 2026; no post has been published yet, and the publishers have only run against test fakes. Meta ads are created paused, and a person starts them.
  • An MCP server and a Telegram console. A Cloudflare Worker with D1 at the edge, and a Google Cloud VM behind Caddy.
  • Estospaces is the first company in it. By 15 November 2026 it had researched a strategy, proposed outreach plays and drafted 5 emails that wait for approval. Nothing had been published, so there are no results to show yet.
EstoGrowth strategy page: who to reach, their pains and where to find them, with unsourced lines marked as guesses
The strategy it researched for Estospaces. Every pain it could not source is marked 'a guess, no quote found'.
A step in EstoGrowth on a phone: review the proposed outreach play for India brokerage owners, status Your step
A step only a person can take, opened on a phone.
export function action(risk: Risk, trust: { level: number }, client: { autonomy: number }, taskType: string, advisoryOk: boolean): Action {
  if (risk === 'block') return 'block'
  if (risk === 'high') return 'client' // founder first, then client
  const lvl = Math.min(trust.level, TASK_CAP[taskType as keyof typeof TASK_CAP] ?? 0, client.autonomy ? 2 : 0, advisoryOk ? 2 : 0)
  return lvl === 0 ? 'founder' : lvl === 1 ? 'veto' : 'auto'
}
EstoGrowth, src/gate.ts, lines 199 to 204, at db1f717 (14 November 2026), verbatim. A blocked post goes nowhere and a high-risk one always goes to a person. Otherwise autonomy is the lowest of four ceilings: earned trust, the cap for the task type, the client's setting and an advisory check. Among the rules that make a post high risk: a number or claim the brand's verified facts do not back, a competitor's name, or being one of the first 5 posts on a platform.
Illustration of the gate above with abstract levels 0 to 2, not app output: a blocked post stops, a high-risk one goes to a person, otherwise the lowest ceiling sets the result (0 founder, 1 veto, 2 auto).

06 Experience

Experience

Work

  • May 2025 to now

    Estospaces, Co-founder and Technical Lead. Led the build end to end and a team of 6 to 8: engineering interns, QA, design and marketing. Interviewed 50+ letting agencies.

  • May 2026 to now

    Zevqo Labs, Founder of my side studio, alongside Estospaces. Website, store-app and CRM set-up for small businesses. One client job so far: a website redesign for Smart Learning Center, a coaching centre in Chennai (in preview). The studio's sites

  • Oct 2024 to Apr 2025

    CourSea Academy, Software Developer. QR certificate verification with Django and Supabase, and a multi-tenant marketing site in React.

  • May 2023 to Sep 2024

    Enkindle Technologies, Bengaluru, Junior Software Developer. Insolvency case management and eVoting apps with React, Flask and AWS.

Earlier

  • 2021

    Feynn Labs and Grroom, Machine Learning Intern

  • 2017 to 2021

    SRM IST, Chennai, B.Pharm

  • Training

    Besant Technologies, full-stack Java bootcamp

Stack

  • Production depth

    Go (Fiber, GORM, pgx; at Estospaces since February 2026), TypeScript and React 19, Flutter, PostgreSQL 16, Supabase, Google Cloud Run and Cloud SQL, Terraform, Docker, GitHub Actions, Playwright, k6, Razorpay

  • Working experience

    Next.js, Node.js and Express, Cloudflare Workers and Pages, Expo and React Native, Chrome extensions

  • Earlier jobs

    Django, Flask, AWS Chalice and Lambda, Gremlin

  • AI-assisted delivery

    Claude Code and Codex under written review rules, LLM guardrails and fact-checking, MCP servers

07 Side projects

Side projects and rapid builds

Also built

Rapid builds

ClientFlow, a client-portal demo for agencies (Vite, React, Clerk, Supabase). JeenieAssistant, a voice receptionist prototype in English, Hindi and Tamil (the AI is mocked). Menopause Tracker, a Flutter app in Android beta.

08 Studio

Zevqo Labs, my side studio

It offers website, store-app and CRM set-up for small businesses. Its one client job so far is a website redesign for Smart Learning Center, a coaching centre in Chennai (in preview). All three sites below are rapid builds.

  • Zevqo Labs studio site, its start-a-project block: need a website, store, CRM or mobile app launched cleanly? Send your business type, the service and the workflow, then WhatsApp or email the studio
    The start-a-project block on the studio's site.

    Studio site

    Next.js 16, Cloudflare Pages

    The studio's own site: what it sets up and how to start a project.

  • Zevqo Labs app demos page: find the right website, CRM, store or mobile app base, with a summary of 166 services, 7 categories and 20 public demos
    The catalogue page and its summary: 166 services, 7 categories, 20 public demos.

    App demos

    Static HTML and CSS

    A searchable catalogue of 166 services and app bases in 7 categories; 20 have public demos. I built the catalogue page. The app bases behind it are purchased code that the studio configures and rebrands for a client; I did not write them.

  • LaunchSite gallery: six sample websites in two rows
    Six of the 20 sample directions: studies, not client work.

    LaunchSite

    Next.js 16, three.js

    My landing-page portfolio: a draggable gallery of 20 live sample sites for schools, clinics, coaches, local services, studios and startups. They are sample directions, studies based on third-party templates, not client work.