Available for work

Maheshwaran A KAI-native
full-stack
engineer.

I build production systems with the parts most demos skip: approval workflow engines, role-based access control and multi-tenant architecture. Dashboards that turn AI output into something an operator can act on. LLM integrations that own the input rather than the output. Full-stack in the ordinary sense — React and TypeScript on the front, FastAPI and Node behind, Postgres and MySQL underneath. At Bipolar Factory I build the retail-analytics command center; before that I founded Statix.pro and took its products 0 → 1.

Salem, Tamil Nadu, India3+ yrs shipping production softwareReact · TypeScript · Python
Illustrated portrait of Maheshwaran A K
~/mahesh.pnghuman in the loop

₹43.4 Cr

Budget processed

1,000+ proposals via my ERP

330+

Active users

across 39 departments

14 mo

In production

multi-tenant ERP

IEEE

Published research

AI recruitment, 2023

01 Experience

Where I've shipped.

Three roles, one through-line: taking systems that are hard to reason about and giving people a surface they can work from.

Current

Sept 2025 — Present

Remote

AI-Native Full-Stack Developer

Bipolar Factory

  • Building the front end and APIs for the enterprise retail-analytics command center, turning the output of the platform’s AI systems into footfall heatmaps, demographics and ANPR views that store operators can act on.
  • Built an end-to-end workflow and scheduling system on FastAPI + PostgreSQL covering recurring, one-off and milestone schedules, TAT cascades, KPI aggregation and manager approval flows.
  • Implemented multi-tenant object storage and evidence upload on S3/MinIO with file validation, size limits, workspace-level isolation and secure reference checks.
  • Shipped React/TypeScript features across video-management and retail-analytics products: ticketing, KPI cards, camera grids, activity search, floor plans and NVR/device integrations.
  • Published a reusable MUI + Recharts dashboard library — KPI cards, footfall heatmaps, widget filters, custom tooltips — as an internal npm package documented in Storybook.
  • Refactored a 100+ file legacy codebase, clearing build and lint failures while standardising on TypeScript and Redux Toolkit.
  • Wired a Next.js analytics dashboard to live backend APIs via Orval-generated typed clients, with shared filtering, SSR fixes, dark mode and responsive charts.

React/TypeScript/Next.js/FastAPI/PostgreSQL/Redux Toolkit/S3 / MinIO/Orval/MUI/Recharts/Storybook

Apr 2023 — Jun 2025

Tamil Nadu

Founder & Full-Stack Engineer

Statix.pro

  • Founded the company and took multiple AI-powered SaaS products 0 → 1, including Guide2Profit, the Statix.pro company site and a research-paper database system.
  • Architected, built and deployed products spanning financial modelling, SEO automation and academic content management.
  • Developed microservice-based backend APIs in Node.js and owned full product cycles — REST API design, cloud deployment and LLM integration.
  • Enforced modular codebases with Git-based CI/CD and OAuth/Auth0 authentication.

Node.js/React/Express/PostgreSQL/Supabase/LLM APIs/Auth0/CI/CD

Jul 2022 — Oct 2022

Tamil Nadu

Web Development Intern

Infoview Technologies

  • Delivered a full-stack blog management system with custom authentication.
  • Applied MVC architecture with JWT sessions and a cross-browser compatible UI.
  • Focused on backend API logic, middleware security and CRUD functionality.

JavaScript/Node.js/Express/JWT/MVC

02 Selected work

Things people actually use.

Five builds that went to production and stayed there. Numbers are real and measured, not estimated.

01

2024 — present

Institutional Finance & Research Platform In production

A multi-tenant college ERP running budgeting, expenditure and research operations for 330+ users across 39 departments — live for 14+ months.

  • A 9-role RBAC approval engine driving multi-stage budget, finance and expenditure workflows — 1,000+ proposals worth ₹43.4 Cr.
  • End-to-end finance workflows across 570+ requests: advances, bill reconciliation, settlements and audit trails.
  • Node.js/Express + MySQL backend and React/TypeScript frontend spanning 15+ business domains.
  • JWT/OAuth authentication with CASL-based permissions for fine-grained, role-specific access.

React/TypeScript/Node.js/Express/MySQL/JWT/OAuth/CASL

02

2026

Land2Build Open source

A construction planner for Indian residential builds: describe a plot and a brief, and it generates a walkable 3D model, a costed material list and a weather-risk view — entirely in the browser.

  • Floor plans are generated, not modelled — binary space partitioning recursively splits the footprint into rooms, with aspect-ratio limits so nothing degenerates into a corridor.
  • Seven house types (bungalow, villa, courtyard, duplex, L-shaped and more) extruded into walkable 3D with React Three Fiber, dimension lines and room labels.
  • An LLM turns a plain-language brief into a typed building config — spans, floors, roof and wall materials — which then drives the geometry and an India-specific cost estimate.
  • Offline-first: projects, materials and costings persist to IndexedDB through Dexie, so the whole tool runs with no backend.
  • Leaflet mapping for the plot and Open-Meteo forecasts for a build-season weather-risk view.

React/TypeScript/Three.js/React Three Fiber/Dexie / IndexedDB/Leaflet/Open-Meteo/Vite

03

2024

Guide2Profit Shelved

A financial-modelling SaaS that lets founders forecast five years out — P&L, break-even and funding requirements — without a spreadsheet.

  • Multi-step form flow feeding backend-triggered financial calculation engines over REST APIs.
  • Dynamic dashboards for 5-year forecasts, P&L statements, break-even analysis and funding estimation.
  • Built on a clean monolithic architecture with functional programming patterns.
  • Automated deploys through Vercel + GitHub CI/CD.
  • Shelved after validation: founders delegate financial modelling rather than doing it — the write-up covers why.

React/Node.js/Express/Supabase/PostgreSQL/Vercel

04

2025

Multi-Agent AI Debate Platform Open source

Two LLM agents argue a topic to resolution — structured argument generation, rebuttal and conflict resolution driven by prompt chaining and function calling.

  • Autonomous multi-agent reasoning loop with structured argument generation and conflict resolution.
  • Function calling and structured outputs through the Nebius AI Studio API.
  • Streamlit frontend streaming debates live on any user-supplied topic.

Python/Streamlit/Nebius AI Studio/Function calling/Prompt chaining

05

2023 — 2025

Statix.pro — Company Site Live

The business website for the company I founded — its public face for AI and digital-marketing work.

  • Designed and built the marketing site end to end: positioning, service pages and enquiry capture.
  • Responsive React front end deployed on Vercel with Git-based CI/CD.

React/Vercel

Looking for a particular kind of work rather than a particular project? What I build

03 Toolkit

What I reach for.

Listed by where it sits in the system, not by how good I claim to be at it. The t-shirt isn't ironic — I ship with coding agents every day, and I've put most of them through real work.

Frontend

ReactNext.jsTypeScriptRedux ToolkitMUIRechartsStorybookStreamlit

Backend

Node.jsExpressPythonFastAPIFlaskDjangoREST APIsOrval

AI & Agents

LangChainLangGraphLangFuseHuggingFaceGPT APIsNebius AI StudioFunction callingGradio

Data

PostgreSQLMySQLMongoDBSupabase

Platform

VercelGitHub ActionsCI/CDS3 / MinIOHostinger

Auth & Access

JWTOAuthAuth0CASLRBAC

AI coding

Claude CodeCursorOpenCodeClineAntigravityCodexGemini CLILovableBolt.newGrokcto.new

Public GitHub

@Maheshwaran325

10 original repos / forks and archives excluded

04 Record

Published, certified, graduated.

Publication

On-Demand Job-Based Recruitment for Organisations Using Artificial Intelligence

Peer-reviewed research on AI-driven recruitment matching, algorithmic scoring and automated candidate evaluation.

IEEE · 25 May 2023

Certification

HuggingFace AI Agents Course

Completed by building and testing an autonomous agent that cleared the GAIA Level 1 benchmark.

HuggingFace · May 2025

Education

B.Tech, Information Technology

K.S. Rangasamy College of Technology.

2019 — 2023 · CGPA 8.67 / 10

05 Questions

Asked and answered.

The things people email to ask before they email to ask anything else.

What kind of work are you available for?

Both full-time engineering roles and freelance or contract projects, remote only. I am currently at Bipolar Factory, so contract work runs alongside that — scoped pieces rather than full-time-equivalent commitments. Email is the fastest way to reach me and I reply within a day.

Do you build multi-tenant systems with role-based access control?

Yes — it is the work I have the most production evidence for. I designed and built a multi-tenant ERP for a 39-department institution: a 9-role RBAC approval engine driving multi-stage budget, finance and expenditure workflows, with CASL-based permissions and JWT/OAuth authentication. It has been live 14+ months, serves 330+ active users, and has processed 1,000+ proposals worth ₹43.4 Cr.

Can you build a custom approval or workflow engine?

Yes, and I have built two in different shapes. The institutional ERP runs a configurable multi-stage approval engine where roles are data rather than code, so a new approval chain is configuration and not a deploy. At Bipolar Factory I built an end-to-end workflow and scheduling system on FastAPI and PostgreSQL covering recurring, one-off and milestone schedules, TAT cascades, KPI aggregation and manager approval flows.

Do you build custom analytics dashboards?

Yes. At Bipolar Factory I build the front end and APIs for an enterprise retail-analytics command center — footfall heatmaps, demographics and ANPR views that store operators act on. I also published the component library underneath it as an internal npm package: MUI and Recharts KPI cards, heatmaps, widget filters and custom tooltips, documented in Storybook.

Do you work with LLMs and AI agents?

Yes, at the product layer — I build the systems around models rather than training the models themselves. In Land2Build an LLM parses a plain-language building brief into a typed config that then drives deterministic geometry, so the generated floor plan can be tested. A multi-agent debate platform runs two LLM agents through structured argument and rebuttal using function calling. I work with LangChain, LangGraph, LangFuse, GPT APIs and Nebius AI Studio.

What is your stack?

React, TypeScript and Next.js on the front; FastAPI, Node and Express behind; PostgreSQL and MySQL underneath. Around that: Redux Toolkit, MUI, Recharts and Storybook; S3 and MinIO for object storage; JWT, OAuth, Auth0 and CASL for authentication and permissions; GitHub Actions for CI/CD.

Where are you based, and do you work remotely?

Salem, Tamil Nadu, India, working India Standard Time (UTC+5:30). Remote only — my current role is remote and I am not looking to relocate. IST gives a full working morning of overlap with Europe and the early hours with US Eastern.

How much production experience do you have?

3+ years shipping production software across three roles: Bipolar Factory today, Statix.pro before it as founder from 2023 to 2025, and an internship at Infoview Technologies. The longest-running thing I have built is the institutional ERP — 14+ months live and still in daily use. I also have peer-reviewed research published by IEEE on AI-driven recruitment matching.

Is any of your work open source?

Yes. Land2Build, the multi-agent debate platform and this site itself are all public on GitHub at @Maheshwaran325. The institutional ERP and the Bipolar Factory products are client and employer work, so those have written case studies rather than repositories.

06 Contact

Let's build something.

Open to full-stack and AI engineering roles. The fastest way to reach me is email — I reply within a day.

LinkedIn

/in/maheshwaranak

GitHub

@Maheshwaran325

X

@gingfreecss325

Based in

Salem, Tamil Nadu

Status

Available for work