TI
Résumé
Iftekher Toufique Imam
Open to new opportunities

Iftekher Toufique Imam

Software Engineer

I build things people actually use — AI photo editors trusted by 10M+ users, NFT platforms on three blockchains, and research cited 79 times. From Metal shaders to smart contracts, I ship across the whole stack.

See my workGet in touch
10M+
Users reached
4+
Years shipping
79
Research citations
7
Products built

Things I've Shipped

Real products, real users — from App Store hits to live blockchain platforms.

Beauty Editor: Edit Face, Body
iOS · AI Photo Editing

Beauty Editor: Edit Face, Body

AI beauty studio loved by 10M+ users worldwide

10M+users4.6★rating15Kratings
  • Photo-realistic virtual makeup — hair recoloring, lipstick, eyelashes, eyeliner, lenses, and teeth whitening rendered on the GPU.
  • Face sculpting tools: resize and reshape face, jaw, lips, nose, and eyes with natural results.
  • Six blur engines (bokeh, hexagonal, disc, zoom, motion, Gaussian) with background, foreground, linear, and circular masking.
  • Full undo-redo manager with layer selection and swapping — desktop-class editing on a phone.
SwiftObjective-CMetal SLCoreML
Color Pop: AI Photo Editor
iOS · AI Photo Editing

Color Pop: AI Photo Editor

One tap turns black-and-white memories into color

4.6★rating78Kratings
  • Built the AI colorization engine that brings any grayscale photo to life.
  • Quick Tools suite: recolor, color-pop, colorize, and double exposure effects.
  • R&D on ML super-resolution, face enhancement, and photo restoration models.
SwiftPyTorchMetal SLCoreML
Wipe: Photo Cleaner
iOS · Utilities

Wipe: Photo Cleaner

Frees gigabytes of storage in one smart tap

4.5★rating450ratings
  • Optimized image-similarity scanning to fly through photo libraries with tens of thousands of items.
  • Designed the ranking algorithm that automatically picks the best shot from a burst of near-duplicates.
SwiftCoreMLVision
Pinky
Web3 · Full Stack

Pinky

Type a prompt, mint an NFT — Web3 in four clicks

3blockchains5 devsteam
  • Led a team of five building a platform that turns generative AI art into NFTs.
  • Wrote and deployed NFT, token, and marketplace smart contracts across Polygon, ASTAR, and TON (FunC).
  • Built the backend and frontend with Next.js, TypeScript, MongoDB, Auth0, and Vercel — wallet to mint in four clicks.
  • Extending to iOS and Android with React Native on the same API layer.
Next.jsTypeScriptSolidityStable DiffusionMongoDB
Pinky Auction
Web3 · Mini App

Pinky Auction

NFT auctions living inside a chat app

Livestatus
  • Gamified NFT giveaway platform built on an auction and investment mechanic.
  • Runs on the TON network with a React/Next.js backend and MongoDB.
  • Live as a LINE Mini App — no install, no wallet setup, straight to bidding.
ReactNext.jsTONMongoDBTelegram Bot
Android · Media

JMXPlayer & JMXPlayer Cast

Every format, every screen — one player

  • Video and audio streaming with casting, playlists, history, and M3U support.
  • In-app browser with automatic media detection and parsing.
  • Built-in ad blocker and cross-app media sharing.
KotlinExoPlayerJavaScriptChromecast
Android · Streaming

Like!

Movie streaming with hardened APK security

  • Movie and TV streaming powered by the TMDB API and custom web scraping.
  • Cut Firebase costs with GitHub-backed content delivery and offline caching.
  • Signature-dependent content encryption to stop tampered APKs cold.
  • Automated the entire CMS pipeline with Node.js.
KotlinNode.jsFirebaseTMDB API

Where I've Worked

Building AI photo tools by day, leading blockchain teams by night.

Software Engineer

Kite Games Studio
Apr 2022 - Present

AI-powered photo editing features shipping to millions of iOS users.

  • Spearheaded AI-powered features for flagship apps with 10M+ combined users — face restoration, super-resolution, and photo-realistic virtual makeup.
  • Built GPU-accelerated face editing tools and image ranking algorithms that lifted user engagement.
  • Automated critical database workflows with scripting, cutting manual data-management work.
  • Hunted down complex performance bugs, delivering measurable gains in app stability and speed.

Blockchain & Backend Developer

Kero.fi Pte Ltd.
Sep 2023 - Oct 2025

Led a five-dev team building an AI-art NFT platform on 3 chains.

R&D Deep Dives

How the magic actually works — real computer-vision pipelines I designed and shipped.

Photo-Realistic Hair Recoloring — processing pipeline

Actual production pipeline — every stage runs on-device.

Photo-Realistic Hair Recoloring

Recoloring hair is easy — until the hair is black, there are two people in the frame, and beards keep sneaking into the mask.

  • Combined a custom tri-map hair model with Face-Parsing outputs to capture fine strands while subtracting beards from the mask.
  • Introduced person segmentation so each person in a group photo can be recolored independently.
  • Black hair carries almost zero signal in LAB or HSV color spaces — solved with multi-layer blending and per-pixel decision making that holds up across every hair color.
  • Shipped with LUT, gradient, and two-color modes plus user controls for coverage, shine, and intensity.
Person SegmentationFace-ParsingTri-map MattingLAB / HSV Blending
Acne Detection & Removal — processing pipeline

Actual production pipeline — every stage runs on-device.

Acne Detection & Removal

Find every blemish on a face, erase it, and keep the skin looking like skin — in real time on a phone.

  • Histogram thresholding proposes candidate acne regions; flood fill strips false positives at the borders, and erosion–dilation maps the affected area.
  • A customized high-pass filter removes blemishes, then blends the result back into the face without the airbrushed look.
  • Implementing this taught me Metal compute shaders — which I then used to parallelize image pipelines across our apps for 10–100x speedups.
Histogram ThresholdingFlood FillHigh-Pass FilteringMetal Compute Shaders
Realistic Lipstick Rendering — processing pipeline

Actual production pipeline — every stage runs on-device.

Realistic Lipstick Rendering

Six lipstick finishes — glossy, matte, metallic, satin, sheer, shimmer — from one light-aware pipeline.

  • Lip masks from face parsing drive a threshold mask that models diffusion and spotlight highlights on the lips.
  • Metallic and shimmer finishes add a procedurally generated glitter mask on top of the light model.
  • The same light-processing core powers every finish except matte, keeping the pipeline fast and consistent.
Face-ParsingLight ModelingProcedural GlitterGPU Blending
Eyebrow Replacement — processing pipeline

Actual production pipeline — every stage runs on-device.

Eyebrow Replacement

Deleting an eyebrow convincingly is harder than drawing one — hair and lashes grow right through it.

  • Segmented out overlapping hair and eyelashes, then inpainted the brow region using only neighboring skin pixels.
  • New brows are blended using both the user’s original brow and the sample brow’s color and lightness data for a natural match.
SegmentationInpaintingColor Transfer

Forehead Landmarks from Scratch

Apple Vision gives you 76 face landmarks — and not one of them is on the forehead.

  • Derived forehead geometry from existing landmarks, but the estimate broke down on side-facing poses.
  • Built a forehead mask by subtracting a landmark-based eyebrow-to-chin mask from the Face-Parsing all-skin mask.
  • Ran Apple's contour model over that mask to extract clean forehead landmarks — robust across face shapes and angles.
Apple VisionLandmark GeometryMask ArithmeticContour Detection

What I Work With

A toolbox that spans from GPU shaders to smart contracts.

Mobile

SwiftObjective-CKotlinReact NativeMetal SL

AI / ML

PyTorchCoreMLComputer VisionStable DiffusionImage Processing

Web3

SolidityFunC (TON)PolygonASTARNFT Marketplaces

Web & Backend

TypeScriptNext.jsReactNode.jsMongoDBFirebase

Credentials

The academic side — cited research and formal training.

Education

B.Sc. in Computer Science and Engineering

Khulna University of Engineering & Technology

2017 - 2022

Thesis: A Novel Multi-Module Approach to Predict Crime Based on Multivariate Spatio-Temporal Data Using Attention and Sequential Fusion Model.

Project: DOC-BLOCK: A Blockchain Based Authentication System for Digital Documents.

IELTS: 7.5 (TRF: 23BD500345IMAI050A)

Publications

79 total citations