Available Immediately • Contract & Full-time

Shipping production AI into real client workflows.

I’m Aman, a forward-deployed full stack AI engineer. I embed with client teams and own the system end to end, from the React frontend to Terraform: computer vision for ITC, a fraud-detection SaaS for Fortza, serverless pipelines on AWS.

See the work Digital CV Book a call

Aman Rajput

Noida, India · remote or relocation

40K+: images a month through the ITC WhatsApp platform
about 50K at peak
84%: mAP@50 on the YOLO model I fine-tuned for ITC
86% precision · 79% recall
~2s: to score a 10,000-row batch in Fortza’s fraud-detection SaaS
ML service answers in 200–300ms

Selected work

Three client systems in production, and what they did.

Each one starts with the problem and the result in plain words. Open “For engineers” for the architecture and stack.

01 · Client: ITC Limited

ITC WhatsApp Image Analysis Platform

The problem: Shop display photos arrive over WhatsApp from across India, and each one has to be checked before a user is paid.
What I built: I owned the platform in production: the YOLO model I fine-tuned on 1,070 annotated shop images, the FastAPI backend, the React admin dashboard and the Terraform-managed AWS database.
The result: Photos are checked automatically and payouts are released after a quick human approval. 40K+ images a month, about 50K at peak.

40K+ Images/Month
84% mAP@50

For engineers

Fine-tuned YOLO on 1,070 custom-annotated shop images across four display classes, reaching 84% mAP@50 (86% precision, 79% recall) on a 270-image validation set.
Restored the admin approval dashboard after RDS connection-pool exhaustion made it unusable at peak load, by resizing the pool, converting code to async and releasing connections before external API calls.
Provisioned a multi-VPC MySQL database on AWS RDS with Terraform.

ReactFastAPIPythonYOLOMySQLAWS RDSTerraform

How it runs

1
WhatsApp photo
Shop images arrive from across India
2
YOLO detection
Fine-tuned model categorizes products, 84% mAP@50
3
Admin approval
A person approves before anything is paid
4
UPI payout
Approved results release the payout

02 · Direct Client Engagement

Fortza Fraud Detection Platform

The problem: Fortza wanted a fraud-detection product its partners could plug into their own point-of-sale and CRM systems.
What I built: I built the Next.js frontend, the FastAPI backend, the inference pipeline and the API-key access layer, working alongside a senior ML engineer.
The result: Fortza sells it as a SaaS to partner organizations. A 10,000-row batch is scored in about two seconds, end to end.

10K Rows in ~2s
ML Service 200–300ms

For engineers

A 10,000-row batch scores in about 2 seconds end to end, with the ML service responding in 200–300ms.
Capped training CSVs at 50,000 rows / 25 MB and test inference at 10,000 rows / 10 MB to guard against oversized inputs.
Enforced ID-based data contracts and batched database writes in groups of 1,000 to avoid PostgreSQL parameter-limit errors.

Next.jsFastAPIPythonPostgreSQLMachine Learning

How it runs

1
Partner system
POS or CRM calls the API with its own key
2
Input guards
Size caps and ID-based data contracts
3
ML scoring
10K-row batch in about 2s, ML service 200–300ms
4
Fraud results
Written to PostgreSQL in batches of 1,000

03 · Feature for Porch Moving Group

Hazard Image Analysis Feature

The problem: Porch Moving Group’s platform needed property photos turned into structured hazard findings.
What I built: I designed, tested and monitored an AWS SQS and Lambda pipeline, provisioned with Terraform, that sends images in batches of 10 to an OpenAI vision model.
The result: The feature went to production in 2–3 days as one part of their larger platform.

Shipped in 2–3 Days
Batches of 10

For engineers

Defined the infrastructure as code with Terraform so the pipeline is reproducible.
Used Claude Code and Kiro for speed while owning the design, test cases and CloudWatch monitoring.

AWS SQSAWS LambdaTerraformOpenAI VisionCloudWatch

How it runs

1
Hazard images
Enter the pipeline from the client platform
2
AWS SQS
Decouples intake from analysis
3
AWS Lambda
Sends batches of 10
4
OpenAI vision
Returns structured JSON findings

More shipped work

Live on Chrome Web Store

TabTrack: Tab Session Manager

Manifest V3 Chrome extension to save, restore and back up tab sessions.

5.0★ on Chrome Web Store • 11 Users

Chrome Extension APIs (MV3)JavaScript (ES6+)JSON

Visit

Live Product

FaceInsight AI

Skin-condition analysis web app with upload validation and an AI Q&A chat.

Live Product • Vision Model + Gemini Chat

Next.jsFastAPIOpenCVGoogle GeminiOpen-source Vision Model

Visit

Open Source on GitHub

Drawvia: Real-Time Collaborative Whiteboard

Shared whiteboard with WebSocket sync across isolated rooms.

Real-Time WebSocket Sync • Turborepo Monorepo

Next.jsReact.jsWebSockets (ws)Node.jsExpress.js

Source

Active R&D (WIP)

Image Intrusion Detection System

Zone-intrusion detection on video streams with automated alerts.

In Development • YOLO + OpenCV

PythonYOLOOpenCVDocker

Live on Vercel

Ghiblify: AI Image Stylizer

Web app that turns user photos into Ghibli-style artwork.

Live on Vercel • Diffusion Models

Next.jsReplicate APIClerk AuthPrismaTailwind CSS

Visit

Built In-House at Urgent IT Solution

RDP Access Control System (MFA)

Multi-factor authentication gateway that keeps RDP blocked until the user verifies.

Firewall-Gated RDP • Time-Limited OTP

PythonFastAPIPowerShellWindows Firewall

Source

Urgent IT Solution • Foreign Clients

LendingHands: P2P Management System

Finance and money-lending SaaS for foreign clients, with loan management and payment flows.

4-Member Team • Foreign Clients

React.jsTailwind CSSPrismaNode.jsMySQL

Source

Across the stack

From the browser to the cloud, one person can carry it.

Four layers I’ve shipped in, each tied to something that’s running.

Frontend & Browser Engineering

Fortza’s Next.js frontend and the published TabTrack extension

Next.js
React.js
TypeScript
JavaScript (ES6+)
Tailwind CSS
Zustand
Chrome Extension APIs (MV3)
WebSockets

Backend & Data

FastAPI behind ITC and Fortza, with the RDS pool fix at peak load

Python
FastAPI
Django
Flask
Node.js & Express
REST APIs
PostgreSQL
MySQL
MongoDB
Prisma ORM
Alembic
Strapi

AI & Computer Vision

YOLO fine-tuned to 84% mAP@50 for ITC

YOLO (Object Detection)
OpenAI Vision API
Google Gemini
Prompt Engineering (Structured JSON)
OpenCV
Replicate API
scikit-learn
Pandas
NumPy
Label Studio
LangGraph

Cloud & DevOps

Terraform-managed RDS, plus an SQS + Lambda pipeline live in 2–3 days

AWS (RDS, SQS, Lambda, CloudWatch, EC2, S3)
Terraform
Docker
GitHub Actions
Vercel
Turborepo
Claude Code
Codex
Kiro

How I work with clients

Built for the forward-deployed seat.

Embed, then own it

I join the client’s team and take responsibility for the system in production, not just the ticket.

ITC: owned the full production lifecycle of the platform.

Find the real cause

When something breaks at peak load, I trace it to the root instead of adding hardware.

Fixed RDS connection-pool exhaustion that froze the approval dashboard, with no added database cost.

Move fast, keep guardrails

I use AI coding tools for speed, and I still own the design, the tests and the monitoring.

Hazard-image pipeline live in 2–3 days, with Terraform and CloudWatch. Fortza inputs capped and validated.

Experience

Where I’ve shipped it.

March 2026 — Present

Noida, Uttar Pradesh, India

Full Stack Developer, Prestanda Consulting

Forward-deployed full-stack and computer-vision delivery for enterprise clients including ITC Limited, Fortza and Porch Moving Group.

Owned the production lifecycle of a WhatsApp-based image analysis platform for ITC Limited, processing 40K+ images a month (about 50K at peak) through automated detection, categorization and an admin dashboard where manual approval gates user payouts.
Restored the admin approval dashboard after RDS connection-pool exhaustion made it unusable at peak load, by resizing the pool, converting code to async and releasing connections before external API calls, with no added database scaling cost.
Shipped an AWS SQS and Lambda pipeline to production in 2–3 days, with Terraform, that sends property hazard images in batches of 10 to an OpenAI vision model for structured-JSON analysis as a feature of Porch Moving Group's larger platform, owning the design, test cases and CloudWatch monitoring.
Built the Next.js frontend, FastAPI backend, inference pipeline and API-key access layer for Fortza's fraud-detection platform alongside a senior ML engineer; a 10,000-row batch scores in about 2 seconds end to end.
Protected Fortza's ML pipeline from oversized inputs and PostgreSQL parameter-limit errors with upload caps, ID-based data contracts and batched writes of 1,000 records.
Fine-tuned YOLO on 1,070 custom-annotated shop images across four display classes, reaching 84% mAP@50 (86% precision, 79% recall).

React · Next.js · FastAPI · Python · YOLO · AWS · Terraform · MySQL · PostgreSQL · OpenAI Vision

July 2025 — March 2026

Noida, Uttar Pradesh, India

Full Stack Developer, Urgent IT Solution

Full-stack delivery of LendingHands, a finance and money-lending SaaS platform for foreign clients.

Owned delivery of the platform from architecture to deployment within a 4-member team, shipping dashboards, loan management flows and payment integrations.
Built an in-house multi-factor authentication gateway for RDP from scratch in FastAPI, with a PowerShell firewall agent that allows only the verified user's IP.
Presented demos and joined client meetings across the Americas and Europe, using technical documentation and interactive prototypes, resulting in repeat business.

React · Node.js · FastAPI · MySQL · PowerShell · REST APIs

Let’s talk

Need an engineer who can own it from first call to production?

I’m available immediately for contract or full-time roles, remote or with relocation.

reachoutaman@outlook.comBook a 15-min call Digital CV

LinkedIn GitHub