Skip to content

Available for work

Akshay Bengani

AI Engineer · Agent systems and technical enablement

I build agent systems for work, and finish the software I need at home.

India · 6 years 9 months · github.com/akshaybengani

Portrait of Akshay Bengani
6+
years shipping software
22
customer engagements delivered
10
applications built and finished
2,661
tests across my three largest codebases

About

What I do, and how I got here

I build and evaluate production agent systems. Over the last year that has meant MCP servers deployed to live customers, RAG pipelines, multi-agent orchestration, and the evaluation harnesses that prove any of it works. Behind that sit six years of shipping mobile and full-stack products, including a PCI-DSS compliant fintech platform used by under-18s across the UAE and Saudi Arabia.

On my own time I finish things: a 112,000 line Rust audio control plane, an on-device agent runtime that never phones home, and a household inventory with 1,295 tests behind it. Most of what I build works with the network unplugged, because that is the constraint I start from rather than a feature I add.

Worth saying plainly, since it shapes everything above. I design these systems, specify them, review every decision, and debug them when they break. The code itself is written with AI, which is also what I do professionally: building agent systems and the harnesses that prove they behave. So ask me why the audio hub owns the clock rather than the client, or why tool restriction belongs at registration rather than on the key. I can defend the architecture. I would not claim to write Rust from memory.

How I work

  • Measure, don't assert. An agent that sounds right and an agent that is right are different claims, and only one survives a test battery.
  • Derive state, never decrement it. A counter that drifts is impossible to audit. A derivation you can recompute is impossible to lose.
  • No network unless the problem needs one. It removes accounts, sync conflicts, outages, and the question of what happens to someone's data.
  • Finish things. Tests are the finish line, not the intention.

Experience

Where I've worked

Six years across regulated fintech, community commerce, developer tooling, and most recently enterprise agent platforms.

  1. Implementation and Technical Enablement

    Jul 2026 to Present

    Mindset AI, as a client of Pratham Softwares

    Enterprise agent platform. Customer-facing delivery across 22 engagements.

    • Built and deployed two production MCP servers powering a live safeguarding alert system, taking it from scoping through deployment, QA, and customer handover.
    • Wrote a probe runner and two 43 case evaluation batteries, and measured detection rates across configurations, so agent behaviour was quantified rather than asserted.
    • Verified eleven live agent plans directly against Firestore, and found and replaced wrong-region crisis helplines before any learner was exposed to them.
    • Diagnosed and fixed two MCP protocol defects that had left both servers unusable by any compliant client, having passed tests that mocked the transport instead of speaking it.
    • Resolved two production escalations on the live platform, from root-cause analysis through merge request to production deployment.
    • Delivered an open-source contribution onboarding agent end to end as a proof of concept.
    • Authored internal enablement tooling and skills, and wrote the workspace routing and architecture documentation for the agent-first platform rebuild.
    MCPPythonFlutterFirestoreBigQueryGCP
  2. Full Stack Senior Developer, internal AI and SDLC

    Jun 2025 to Jul 2026

    Pratham Softwares (PSI)

    Internal innovation: AI automation, developer tooling, and full-stack delivery.

    • Built Aegis AI (v3.3.9), a containerized agent running on GitLab CI/CD schedules that scans the issue tracker for labelled work, dispatches AI agents to analyse or fix it, and reports back as comments and merge requests. Added Anthropic and OpenHands provider support and the Docker packaging.
    • Built an AI SDLC playbook and Claude Code plugin for an internal .NET 8, EF Core, PostgreSQL, and React 18 portal, with six specialised subagents, slash commands, hooks, and a companion knowledge repository.
    • Authored five agentic AI workshops for the engineering organisation, including a RAG support agent on LangChain and FAISS with configurable embedding and model providers.
    • Delivered full-stack applications in React, FastAPI, and PostgreSQL, containerized to Azure Container Apps, and designed multi-stage CI/CD pipelines with error handling and automated reporting across five platforms.
    • Cleared Azure AI-900, and was recognised as Emerging Employee of the Year 2025.
    PythonFastAPIReactDockerAzureGitLab CILangChain
  3. Frontend Developer

    Mar 2025 to May 2025

    Sheldom Labs · Gomegle

    Discover people, chat in real time, pick a meetup spot, and navigate there. Shut down when funding ran out.

    • Built the Flutter application with real-time chat over WebSockets, Mapbox location tracking, and a gamified interaction layer.
    • Worked to a modular BLoC and Cubit architecture with reusable widgets, custom animations, and multi-language support.
    FlutterDartFirebaseWebSocketsMapbox
  4. Senior Mobile Application Developer

    Sep 2021 to Dec 2024

    Cashee Tech FZE

    Banking and financial literacy for ages 8 to 18 across the UAE and Saudi Arabia. Real cards, real money, real compliance.

    • Led end-to-end Flutter development, building core financial features to PCI-DSS compliance with root and VPN detection.
    • Implemented Apple Wallet card provisioning and native bridges for third-party SDKs.
    • Built gamified educational modules, multi-language UI, and analytics through Countly and CleverTap, and tuned performance across several markets.
    signups in 3 months
    7,500+

    signups in 3 months

    cards issued
    800+

    cards issued

    UX Design Awards nominee
    2024

    UX Design Awards nominee

    FlutterDartiOS SDKPCI-DSSApple Wallet
  5. Frontend Developer

    Feb 2020 to Sep 2021

    Rirev · Sync, built for Swiggy

    Invite-only communities buying from local vendors. Became the groundwork for Swiggy Mini.

    • Built community joining, shopping, and vendor chat in Flutter.
    • Designed in-app logic for supporting multiple communities, and contributed to backend integration over GraphQL and Hasura.
    FlutterDartGraphQLHasuraFirebasePostgreSQL

Jan 2025 to Feb 2025
A deliberate break after Cashee. I travelled, rested, and went deep on AI. It is the reason the rest of this reads the way it does.

Projects

The ones I'd want read first

Large, finished systems built outside work. All of them are private while they settle, so the write-ups here are the way in.

Also built

One problem each, followed to a finished app

Smaller than the three above, and every one of them started with a specific irritation in my own week.

Skills

What I actually work in

Ordered by what I reach for now rather than what I once read a book about.

AI and agents

Building agent systems, and proving they behave.

AnthropicLangChainOllamaFastAPIPython

Languages I work in daily

Six years of professional work, most of it mobile.

DartPythonTypeScriptJavaScriptJava

Systems I architected and shipped

Specified, reviewed and debugged by me, written with AI. I can defend the design decisions, not recite the syntax.

RustSwiftElectronGo

Mobile and desktop

Six years of it, including a regulated fintech platform.

FlutterAndroidAppleReact

Backend and data

Services, storage, and the queries behind them.

PostgreSQLRedisFirebaseGraphQLPostman

Infrastructure

Getting it to production and keeping it there, including on hardware I own.

DockerProxmoxpfSenseKubernetesCloudflareLinuxNginxGoogle Cloud

Delivery

Multi-stage pipelines with error handling and reporting, across five platforms.

GitHubGitLabJenkinsSonarQubeJira

Beyond the job

There's a whole other page about who I actually am

The Proxmox node and the wall of cable behind it. A projector automation that presses forty buttons so I don't have to. A house on solar. Kubernetes, currently being learned the hard way. And the photos.

Read the personal page
The homelab wall: shelves of boards, routers, drives and a great deal of cable

Credentials

Education, certifications, and awards

Education

  • MCA, Master of Computer Application

    Jaipur National University, Rajasthan

    2020 to 2022

  • BCA, Bachelor of Computer Application

    JECRC University, Jaipur

    2016 to 2019

Certifications

  • AI-900, Microsoft Azure AI FundamentalsCertified
  • AI-102, Azure AI Engineer AssociateIn progress
  • Full Stack AI EngineerTraining
  • Google Prompting EssentialsTraining
  • CompTIA Security+Training
  • CISCO Networking Basics SpecializationTraining

Awards

  • Emerging Employee of the Year 2025

    Pratham Softwares

  • Adhack 5.0 Winner

    AdHoc Networks Infotech, Jul 2019

Full resume (PDF)
{
  "created_at": "2026-09-01T11:40:19+05:30",
  "updated_at": "2026-09-21T18:51:29+05:30"
}