AI Engineer · Multi-Agent · Local LLMs

Amman
Faisal

I build AI systems end to end — multi-agent orchestration, RAG pipelines, and computer vision systems. From prototype to production-ready code.

GenAIMulti-AgentRAGLangGraphOllamaComputer Vision

Resume ↗

AI & ML Projects

Open Source Repos

Technologies

Years of Experience

Core Expertise

Multi-Agent Orchestration88%

Production RAG Pipelines90%

Ollama / Local LLMs90%

Computer Vision82%

Python / FastAPI95%

Scroll to explore

01 · Profile

About Me

Amman Faisal

AI / ML Engineer

EducationFAST NUCES ‘26

LocationIslamabad, PK

I'm an AI/ML engineer and a Data Science graduate of FAST NUCES, Islamabad.

I build AI systems end to end — from multi-agent orchestration and RAG pipelines to production computer vision. I've shipped real systems at the KDD Lab at FAST NUCES (LLM fine-tuning research). Not side projects — production-grade builds.

I'm a strong advocate for local-first AI — running systems on Ollama and OpenRouter to keep data private and costs near zero.

amman.config

const amman = {
location: "Islamabad, PK", status: "Open to work — FAST NUCES graduate",
openTo: ["work", "freelance", "collaborations"]
}

02 · Timeline

Where I've Built

ML Engineer, RL Environments

Tensium

Sep 2026 – PresentActive

Design long-horizon, multi-step software-engineering tasks used as RL environments for frontier-model post-training; 20+ tasks submitted and accepted with a 100% approval rate.

Engineer each task to sit beyond current frontier capability, so leading coding agents (most recently Qwen3.8-Max) fail it, producing high-signal training data on real ML engineering gaps: data leakage, broken evaluation, silent data errors, and slow pipelines that must get faster without changing their output.

Build sealed Docker environments with reproducible setup, reference solutions, and deterministic test-based verifiers for reward; run agent rollouts through mini-SWE-agent and Harbor harnesses to validate difficulty and grading.

Analyze agent trajectories to pinpoint failure modes and harden tasks with negative controls against shortcut solutions and reward hacking before submission.

Product Owner, Gyraq & Founder, IdeaVault

Gyraq

Dec 2025 – PresentActive

Flamingo: own a consent-first AI copilot for software agencies that turns client scoping calls into quote-cited requirements, suggests follow-up questions ranked by scope and risk, and drafts summaries, UI sketches, and effort estimates. In private pilot.

Reach: own a real-time team collaboration platform used by 50+ people with zero downtime, offline-first sync, role-based onboarding, and a tamper-evident activity log.

IdeaVault: founded and built an AI research notepad app (Flutter, Android and web) that turns business ideas into reports using Tavily and Gemini with BM25-reranked news, stored in the user's own Google Drive. In control-group testing.

AI Engineer

Zamooq

Aug 2025 – PresentActive

Built the AI layer across Zamooq's products: ZNXCare (healthcare), ZNXErp (ERP), and ZNXePOS (point-of-sale), covering multilingual voice agents, LLM chatbots, and RAG systems serving 3,000+ enterprise users.

Designed real-time voice AI pipelines combining streaming STT/TTS (Deepgram, Cartesia, ElevenLabs), LLM reasoning (Claude, Gemini), and interruption handling over LiveKit and SIP. Cut agent response time from 8.0s to 0.93s and brought end-to-end latency under 800ms through inference tuning and model selection.

Built RAG pipelines on Qdrant that ground agent answers in business knowledge to reduce hallucinations, and orchestrated multiple LLMs (Ollama, OpenRouter, Claude, Gemini) with cost- and latency-aware routing.

Worked with clients on the product side to scope where AI adds value and define agent behavior. Shipped compliance-friendly AI workflows iteratively from real usage, cutting manual work by 75%.

AI Engineer

Cyberlytics Lab

Aug 2025 – Jul 2026Cleared

Architected CODE-AI, an end-to-end software security framework for vulnerability detection, exploit reasoning, secure code generation, and patch validation.

Built a semantic retrieval pipeline indexing 168,960+ NVD records in ChromaDB with Sentence Transformers; temporal evaluations showed legacy local LLMs (Ollama) correctly identifying post-2025 CVEs from RAG context alone.

Engineered a dual-model inference architecture (DeepSeek-R1 & Llama 3.1) that cross-validates patches to eliminate hallucinations during automated code repair.

Evaluated across 300 test cases with 4 testing frameworks: 86.7% true positive detection rate (260/300).

AI Associate (Internship)

Cyberlytics Lab

Jun 2025 – Aug 2025Cleared

Contributed to initial AI research and development pipelines, laying the technical foundation for context-aware vulnerability retrieval.

Fine-tuned Transformer-based classification models on email text, visual attachments, and header metadata to detect and categorize phishing and cyber attacks.

AI Engineer Intern

Envysion AI

Mar 2025 – May 2025Cleared

Worked on medical AI for ECG and clinical machinery data, predicting critical diseases in premature infants at 88%+ accuracy (PyTorch, computer vision).

Generative AI Intern

KDD Lab, FAST NUCES

Jul 2024 – Aug 2024Cleared

Fine-tuned Mistral-7B with QLoRA for NL-to-Pandas translation, reaching ~80% functional accuracy and +15% structural correctness.

Built a CSV query assistant with schema-aware generation and safety checks.

Co-created a 76.7K-entry text-to-Pandas dataset and published the model and dataset on Hugging Face.

Freelance AI Engineer

Fiverr

Sep 2023 – Feb 2026Cleared

Delivered freelance AI projects alongside my degree: the HVACServicesAI home-service platform (Claude, pgvector, BullMQ), a production LiveKit voice agent for cold calling and website support, a Playwright lead-generation and reputation-intelligence platform with WhatsApp outreach, and an offline flour mill billing system.

Education

BS Data Science

FAST National University, Islamabad

Graduated June 2026Cleared

03 · Builds

Projects I've Built

View All

Products

F1Product Owner · AI Meeting Copilot

Flamingo — Consent-First AI Meeting Copilot

Product

Product owner — in a controlled pilot

TypeScriptNode.js 22Next.js 16React 19Auth.jsNeon PostgresDrizzle ORMpgvectorVercelGitHub Actions

F2Product Owner · Collaboration App

Reach — Real-Time Team Collaboration

Product

Product owner — shared boards for company teams

VercelCloudflare WorkersResend

F3Creator · Idea Research App

IdeaVault — Idea Capture & AI Research

In Dev

Research

Privacy-preserving machine unlearning

PythonPyTorchFederated LearningNeural NetworksPrivacy ML

15NLP

Text Humanizer

In Dev

Rewrites AI-generated text to read naturally

PythonLLMNLP

Building...

16Voice AI

Sara — WhatsApp Urdu Voice Agent

Active

Voice & chat customer support / reception for Urdu speakers

FastAPIGroq (Llama 3.3 70B)faster-whisperAzure SpeechChromaDBElectronReact

Code

17Computer Vision

Computer Vision AR Toolkit

In Dev

MediaPipe & InsightFace experiments

PythonMediaPipeInsightFaceOpenCVNode.js/Express

Code

18Automation / Data

AI Lead Generation & Reputation Intelligence Platform

Active

Freelance client deployment — NDA

PythonPlaywrightFastAPIOllama/LLMWhatsApp Cloud APIDocker

19Voice AI

AI Voice Agent — Cold Calling & Website Support

Active

Freelance client deployment via LiveKit

LiveKitVoice AIReal-time AudioSTT/TTS PipelineLLM

20Computer Vision

Video Upscaler

In Dev

AI-powered video resolution enhancement

PythonPyTorchComputer VisionSuper-Resolution

Building...

04 · Arsenal

Tools & Stack

LangChain

90%

LangGraph

85%

OpenAI API

92%

Anthropic API

88%

HuggingFace

85%

Ollama (Local LLMs)

90%

OpenRouter

82%

PyTorch

80%

QLoRA Fine-tuning

78%

RAG Pipelines

90%

Vector DBs (ChromaDB)

88%

Prompt Engineering

92%

Also experienced with

Mistral-7BQLoRAPEFTStreamlitChromaDBn8nDeepSeek-R1LLaMA 3.1WhisperEdge TTSGitFastAPIPandasHugging FaceJupyterNext.jsTypeScriptOllamaPyTorchLangGraphOpenRouterCrewAILlamaIndexRAGAutogen

05 · Connect

Let's Build

Open to work — full-time roles, freelance projects, and research collaborations. Building something with LLMs, agents, or computer vision? Let's talk.

khawajaamman@gmail.com

GitHub

github.com/Stradok

amman-faisal

Islamabad, PK · Open to Remote Worldwide

AmmanFaisal

About Me

Amman Faisal

Where I've Built

ML Engineer, RL Environments

Product Owner, Gyraq & Founder, IdeaVault

AI Engineer

AI Engineer

AI Associate (Internship)

AI Engineer Intern

Generative AI Intern

Freelance AI Engineer

BS Data Science

Projects I've Built

Flamingo — Consent-First AI Meeting Copilot

Reach — Real-Time Team Collaboration

IdeaVault — Idea Capture & AI Research

University Management System — Database Design

Energy Forecasting & Machine Learning Analysis

Graph Algorithms & Problem Decomposition

Big Data Analytics with Apache Spark

Qalb — Largest State-of-the-Art Urdu LLM

Self-Hosted Urdu Voice Agents

CODE-AI — Vulnerability Detection Pipeline

Content Moderation AI Audit — Bias, Robustness & Safety

JAN — Joint Autonomous Neural Agent

LaunchMind — Autonomous Startup Launcher

Flour Mill Management System

Cody Local — Local AI Coding Assistant

TEXT PANDAS — Tabular RAG Engine

Federated Unlearning in Neural Networks

Text Humanizer

Sara — WhatsApp Urdu Voice Agent

Computer Vision AR Toolkit

AI Lead Generation & Reputation Intelligence Platform

AI Voice Agent — Cold Calling & Website Support

Video Upscaler

Tools & Stack

Let's Build

Amman
Faisal