Ismail Soomro

# Technical Skills

Core Capabilities

Intelligence

RAG chatbots and private AI assistants — not generic API wrappers.

Agentic Workflows RAG Pipelines Vector Search Embeddings Fine-Tuned Models Transformers Quantization RAG Vector Databases

Computer Vision

Production-grade vision pipelines — detection, face matching and identity verification at web scale.

Transfer Learning Image Segmentation NeRFs VLMs Autonomous Driving Spatial Computing & XR Generative Vision & Diffusion Models scikit-learn

TensorFlow

Generative & Voice AI

Voice cloning and generative audio/image pipelines, from a single sample to production output.

Autonomous Voice Agents & Call Centers Healthcare Dictation & Scribes Automotive & IoT Voice Controls Voice Cloning GANs

Automation

Multi-agent systems with tool calling, LangGraph & n8n orchestration, and ecosystem integrations that take action, not just trigger alerts.

Multi-Agent Systems Tool Calling LangGraph n8n CRM Integrations Stripe Twilio Slack

Full-Stack Products

Full-stack, AI-native SaaS platforms built from concept to production — modern frontends, robust backends and scalable databases.

React / Next.js

FastAPI

Node.js PostgreSQL MongoDB

SQL

Cloud & Infra (AWS & Azure)

Cloud-native deployment — containers, managed ML infra and observability baked in.

AWS SageMaker Lambda EC2 S3 DynamoDB ECS Fargate CloudWatch

Azure

Docker

# Projects

production AI systems that shipped, not notebooks that ran once

Flagship project · Voice AI

VoiceForge — AI Voice Cloning & Custom Audio Generation

An agentic AI service for voice cloning and audio generation from a single sample — autonomously handling clone regeneration, multi-output creation and resource optimisation. Integrates ElevenLabs' voice models with a FastAPI backend on AWS S3 + DynamoDB. If a user's clone goes missing, the backend regenerates it on the fly from the original sample, transparently to the user.

Self-healing clones — missing voice clones are regenerated automatically from the stored sample

Thread pooling & semaphores handle simultaneous requests without blocking

Outputs auto-expire after 10 minutes to keep the system memory-optimised at scale

VoiceForge AI voice cloning concept visual — Illustrative concept art — not a literal product screenshot.

FastAPIElevenLabs APIAWS S3DynamoDBAgentic AI

Face Matching Engine Service

A production system that finds visually similar faces on the web from a single uploaded photo, ranking results with learned embeddings and cosine similarity. Combines YOLOv8 person detection with InsightFace face embeddings behind a FastAPI service with background jobs, concurrency controls and fault-tolerant downloads, deployed on AWS ECS Fargate.

~91% top-1 precision@10 on an internal validation set

>98% download success rate with retry and circuit-breaker logic

~1.8s median latency for 20 candidates on a T4 GPU

FastAPIYOLOv8InsightFaceAWS ECSS3

AI Juggling Detection Service

A high-performance FastAPI microservice that detects and counts juggling patterns in video, supporting both direct uploads and S3-based processing. Built on YOLOv11 with asynchronous task management, configurable resource limits and automatic cleanup — designed for sports analytics and automated video analysis at scale.

Async, non-blocking processing for large video files

Full HTTP error taxonomy, retry logic and health-check monitoring

FastAPIYOLOv11asyncioAWS S3

FaceScan — Custom Face Detection Engine

A domain-specific face verification service built entirely from scratch, without pre-trained models. Trained on curated image datasets to encode facial features, verify a submitted image against known identities, and return structured metadata — name, age and other attributes — with a confidence score, backed by MySQL.

Custom-trained feature encoding, no third-party face-recognition API

Structured, JSON metadata output for easy integration into other pipelines

Custom MLFace VerificationMySQL

# About