Abu is available for hire
Hire AbuAbu Bakar Ilyas
Verified Expert in Engineering
AI Engineer and Developer
Karachi, Pakistan
Toptal member since September 10, 2026
Bio
Abu is a principal AI engineer with 7+ years of experience in ML and LLMs. He builds agentic systems and designs harnesses that empower agents to manage entire workflows in production, adhering to enterprise standards like validation and defensive coding for Fortune-class banking clients and startups. With a master's degree in computer science from Georgia Tech, Abu placed first among 18,000 participants in a Kaggle competition.
Portfolio
Resilia
Large Language Models (LLMs), Model Context Protocol (MCP), Containerization...
Niceone
Agentic AI, Deep Learning, Forecasting, Machine Learning, Data Engineering...
360factors
Large Language Models (LLMs), Kubernetes...
Experience
- Python - 6 years
- Statistical Forecasting - 6 years
- AI Engineering - 6 years
- Forecasting - 6 years
- Reinforcement Learning - 3 years
- Model Context Protocol (MCP) - 3 years
- LLM Fine-tuning - 3 years
- Agentic AI - 3 years
Preferred Environment
Azure, Python, PyTorch, Transformers, TypeScript, AWS IoT, AI Engineering
The most amazing...
...thing I've built is KAIA, a multi-agent compliance system that now runs in production for 8 US banks, outperforming ChatGPT and Claude in terms of accuracy.
Work Experience
Principal AI Engineer/Consultant
2026 - PRESENT
Resilia
- Built an automated ad-generation and research system where agents managed end-to-end workflows, fine-tuning an LLM with reinforcement learning as a self-verifying harness that continuously evaluated the pipeline and auto-promoted the best scaling.
- Designed reusable skills, context, and MCP tools, boosting a lean team's delivery capacity by multiples. Reduced needed interventions, enabling closed-loop operations and parallel processing across advanced coding agents.
- Engineered a fully automated, containerized pipeline that generates complete ad videos from only a concept and product description, achieving higher visual consistency and quality than leading tools such as Higgsfield.
Technologies: Large Language Models (LLMs), Model Context Protocol (MCP), Containerization, Data Pipelines, Communication, Model Development, Stakeholder Management, XGBoost, Artificial Intelligence (AI), Generative Artificial Intelligence (GenAI), LangGraph, Model Tuning, AI Modeling, JavaScript, Attention to Detail, Screeners, Written Communication, Code Review, Team Leadership, Interviewing, Team Management, Full-stack, API Integration, Agentic RAG Systems, Minimum Viable Product (MVP), Claude API, OpenAI, Booking Systems, Localization, Mobile Apps, WhatsApp API, OpenAI API, User Experience (UX), User Interface (UI), Stripe Payments, SQL, Optical Character Recognition (OCR), Data Analysis, Conversational AI, PostgreSQL, Node.js, Supabase, React, FastAPI, Amazon Web Services (AWS), Solution Architecture, AI Agents, Multi-agent Systems, Answer Engine Optimization (AEO), Gemini API, Prompt Engineering, ChatGPT API, SEO Tools, Marketing, Technical Architecture, System Architecture, AWS IoT, AI Integration, Fractional CTO, CTO, SaaS, Cloud Architecture, PostgreSQL 10, Next.js, System Architecture Design, Software System Architecture Development, Technical Leadership, Scalability, Software Engineering, Agentic AI Systems, LangChain, API Design, Docker, Google Cloud Platform (GCP), Performance, Performance Optimization, Vector Search, Microsoft Azure, APIs, Testing, Azure Cloud Services, Amazon Bedrock AgentCore, AI Architecture, Context Engineering, RAG Systems, Neo4j, Ontologies, Redis, Temporal, Knowledge Graphs, Databricks, Data Modeling, SPARQL, Business Analysis, Semantics, Graph Databases, Spark, Data Build Tool (dbt), ETL, Azure OpenAI Service, Microsoft Copilot, Microsoft Copilot Studio, Technology, Strategy, Claude Code, PyTorch, Natural Language Processing (NLP), TensorFlow, Fine-tuning, Security Audits, Data Security, Startups, SOC 2, Document Processing, Hugging Face, CrewAI, Gemini, Google ADK, Retool, Agentic Frameworks, RAG Architecture, Codex, Cloud Platforms
Principal AI Engineer/Consultant
2025 - PRESENT
Niceone
- Provided consulting in machine learning and demand forecasting, developing algorithms to enhance the accuracy of future demand predictions for a catalog of around 18,000 items.
- Developed an ensemble of forecasting algorithms that predicts demand 12 months ahead at the individual retail-store level, outperforming the company's previous human-driven, Excel-based forecasts across all product categories.
- Automated the forecasting pipeline so that ongoing effort is limited to monitoring data drift and making minor adjustments, freeing the planning team from manual forecasting.
Technologies: Agentic AI, Deep Learning, Forecasting, Machine Learning, Data Engineering, Demand Forecasting, Data Science, Time Series Forecasting, Data Pipelines, Model Evaluation, Communication, Model Development, Feature Engineering, Monitoring, Stakeholder Management, XGBoost, Model Deployment, Linear Regression, Artificial Intelligence (AI), Generative Artificial Intelligence (GenAI), Azure, Model Tuning, AI Modeling, Attention to Detail, Screeners, Written Communication, Code Review, Team Leadership, Interviewing, Team Management, Full-stack, API Integration, Agentic RAG Systems, Minimum Viable Product (MVP), Claude API, OpenAI, Booking Systems, Localization, Mobile Apps, WhatsApp API, OpenAI API, User Experience (UX), User Interface (UI), Stripe Payments, SQL, Data Analysis, Conversational AI, Large Language Models (LLMs), PostgreSQL, Node.js, Supabase, React, FastAPI, Amazon Web Services (AWS), Solution Architecture, AI Agents, Multi-agent Systems, Gemini API, ChatGPT API, Technical Architecture, System Architecture, AWS IoT, AI Integration, Fractional CTO, CTO, SaaS, Cloud Architecture, PostgreSQL 10, Next.js, System Architecture Design, Software System Architecture Development, Technical Leadership, Scalability, Software Engineering, Agentic AI Systems, LangChain, API Design, Docker, Google Cloud Platform (GCP), Performance, Performance Optimization, Vector Search, Microsoft Azure, APIs, Testing, Azure Cloud Services, Amazon Bedrock AgentCore, AI Architecture, Context Engineering, RAG Systems, Neo4j, Ontologies, Redis, Temporal, Databricks, Data Modeling, SPARQL, Business Analysis, Semantics, Graph Databases, Spark, Apache Airflow, Data Build Tool (dbt), ETL, Azure OpenAI Service, Microsoft Copilot, Microsoft Copilot Studio, Technology, Strategy, Claude Code, Model Context Protocol (MCP), PyTorch, Natural Language Processing (NLP), TensorFlow, Fine-tuning, Security Audits, Data Security, SOC 2, Document Processing, Hugging Face, Agentic Frameworks, RAG Architecture, Cloud Platforms
Principal AI Engineer
2024 - 2026
360factors
- Led a senior, multidisciplinary team of 6-7 members in front-end, back-end, data, and ML development. Scoped work, reviewed approaches and PRs against a shared system model, and upheld enterprise quality standards before merging.
- Built the company's AI-powered agentic platform, KAIA, from scratch. Since its launch, KAIA has attracted approximately 7 new banking and credit union clients. The answers and automation provided by KAIA surpass those of ChatGPT, Claude, and others.
- Architected and shipped an AI-powered compliance expert system on an agentic RAG architecture with specialized LLMs, implementing defensive coding, validation, edge-case handling, and automated evaluations throughout.
- Got the Best Employee Award of the year in 2025 due to my extensive work on the KAIA chatbot and complete ownership of the product.
- Shipped enterprise features full-stack from reviewed plans/specifications through agent-assisted execution to merged production releases fast enough to reflect product and customer feedback within the day.
- Fine-tuned a Llama-3 agent for reliable NL-to-SQL conversion with robust guardrails to prevent hallucinations.
- Integrated Microsoft Fabric for conversational dashboard exploration.
- Managed products end to end as a one-person team, handling containerization with Docker and Kubernetes, training and inference, CI/CD delivery, model versioning, monitoring, and automated retraining using MLflow.
- Established services, shipped them, and operated them without handoff.
Technologies: Large Language Models (LLMs), Kubernetes, Reinforcement Learning from Human Feedback (RLHF), RAG Pipelines, Azure, Data Engineering, Demand Forecasting, Data Science, Time Series Forecasting, Data Pipelines, Model Evaluation, Communication, Model Development, Feature Engineering, Monitoring, XGBoost, Model Deployment, Linear Regression, Artificial Intelligence (AI), Generative Artificial Intelligence (GenAI), LangGraph, Model Tuning, AI Modeling, JavaScript, Attention to Detail, Screeners, Written Communication, Code Review, Team Leadership, Interviewing, Team Management, Full-stack, API Integration, Agentic RAG Systems, Minimum Viable Product (MVP), Claude API, OpenAI, Booking Systems, Localization, OpenAI API, User Experience (UX), User Interface (UI), Stripe Payments, SQL, Optical Character Recognition (OCR), Data Analysis, Conversational AI, PostgreSQL, Node.js, Supabase, React, FastAPI, Amazon Web Services (AWS), Solution Architecture, AI Agents, Multi-agent Systems, Answer Engine Optimization (AEO), Gemini API, Prompt Engineering, ChatGPT API, Audits, Auditing, Technical Architecture, System Architecture, AWS IoT, AI Integration, Fractional CTO, CTO, SaaS, Cloud Architecture, PostgreSQL 10, System Architecture Design, Software System Architecture Development, Technical Leadership, Scalability, Software Engineering, Agentic AI Systems, LangChain, API Design, Docker, Google Cloud Platform (GCP), Performance, Performance Optimization, Vector Search, Microsoft Azure, APIs, Testing, Amazon Bedrock AgentCore, Datadog, AI Architecture, Context Engineering, RAG Systems, Neo4j, Ontologies, Redis, Temporal, Knowledge Graphs, Data Modeling, SPARQL, Business Analysis, Semantics, Graph Databases, Spark, Apache Airflow, Data Build Tool (dbt), ETL, Azure OpenAI Service, Microsoft Copilot, Microsoft Copilot Studio, Technology, Strategy, Claude Code, Model Context Protocol (MCP), PyTorch, Natural Language Processing (NLP), TensorFlow, Fine-tuning, On-premise, Security Audits, Data Security, Startups, SOC 2, Document Processing, Legal Technology (Legaltech), Hugging Face, CrewAI, Gemini, Google ADK, Retool, Agentic Frameworks, RAG Architecture, Codex, Cloud Platforms
Data Scientist
2022 - 2024
Afiniti
- Used Microsoft AutoGen to build autonomous multi-agent systems with an orchestrator that instantiated sub-agents for tasks like news retrieval, research-paper analysis, GUI automation with OmniParser and Qwen, with multi-turn evaluation harnesses.
- Led migration of Afiniti's core ML pipeline from R to Python and overhauled its graph-based call-center pairing pipeline with automated grid search, fold-wise validation, automated deployment, and diagnostic tooling reused across the team.
- Increased gain over naive methods by approximately 10%, consistently delivering monthly revenue exceeding the $300,000 target.
- Received the best Rising Star award from the company in the AI and data team.
Technologies: R, Python, Bayesian Networks, Machine Learning, Deep Learning, Data Engineering, Demand Forecasting, Data Science, Time Series Forecasting, Data Pipelines, Model Evaluation, Communication, Model Development, Feature Engineering, Monitoring, XGBoost, Model Deployment, Linear Regression, Artificial Intelligence (AI), Model Tuning, AI Modeling, SQL, Optical Character Recognition (OCR), Data Analysis, FastAPI, Amazon Web Services (AWS), Software Engineering, LangChain, API Design, Docker, Google Cloud Platform (GCP), Performance, Performance Optimization, Vector Search, Microsoft Azure, APIs, Testing, Amplitude, Datadog, RAG Systems, Neo4j, Ontologies, Redis, Temporal, Knowledge Graphs, Spark, Apache Airflow, Data Build Tool (dbt), ETL, PyTorch, Natural Language Processing (NLP), TensorFlow, Fine-tuning, On-premise, Security Audits, Data Security, Document Processing, Hugging Face
ML Engineer
2020 - 2022
The AI Systems Pvt
- Built ML, probabilistic, and deep learning models, including custom stacked LSTMs and text-data encodings, for time-series forecasting and anomaly detection.
- Delivered demand forecasting for Medusa Distributions (USA) at 90% accuracy across 8,000+ SKUs.
- Automated supply-chain pipeline for Arcelik, achieving over 20% accuracy.
- Developed ML pipeline for Engro, surpassing prior forecasts by 30%.
Technologies: Machine Learning Operations (MLOps), Machine Learning, Deep Learning, Statistical Modeling, Data Engineering, Demand Forecasting, Data Science, Time Series Forecasting, Data Pipelines, Model Evaluation, Model Development, Feature Engineering, Monitoring, XGBoost, Model Deployment, Linear Regression, Artificial Intelligence (AI), Model Tuning, Embedded Systems, AI Modeling, Interviewing, Team Management, Full-stack, API Integration, Agentic RAG Systems, Minimum Viable Product (MVP), SQL, Optical Character Recognition (OCR), Data Analysis, Conversational AI, FastAPI, SaaS, Software Engineering, Docker, Google Cloud Platform (GCP), Performance, Performance Optimization, Vector Search, Microsoft Azure, APIs, Testing, Amplitude, Spark, Apache Airflow, Data Build Tool (dbt), ETL, Snowflake, PyTorch, Natural Language Processing (NLP), TensorFlow, Fine-tuning, On-premise, Security Audits, Data Security, Startups, Document Processing, Legal Technology (Legaltech), Hugging Face
Experience
KAIA | My Compliance Expert
https://ask-kaia.360factors.com/A Java and Python-based enterprise application that serves banking and insurance clients.
I worked on KAIA from scratch, from ideation through launch. KAIA differs from ChatGPT and Claude in that it is trained exclusively on federal and state laws and examiner guidelines. KAIA is an agentic orchestration system that solves every aspect of the compliance domains. It is a secure system that uses self-deployed and fine-tuned models on our GPUS. Due to the security requirements of the financial domain, none of the clients' data could be sent to third parties, so it was necessary to store all data in our own environment.
I designed KAIA's agentic architecture and built the AI-based orchestration and logic. This involved fine-tuning and using RLHF to align KAIA's answers with those in the banking domain. I evaluated the accuracy and designed the ingestion pipelines to fetch weekly data from government websites. I also integrated the entire KAIA with the front end, managed the front end and DevOps teams, and researched and provided appropriate APIs and deployment guidelines to the front end and DevOps teams.
I worked on KAIA from scratch, from ideation through launch. KAIA differs from ChatGPT and Claude in that it is trained exclusively on federal and state laws and examiner guidelines. KAIA is an agentic orchestration system that solves every aspect of the compliance domains. It is a secure system that uses self-deployed and fine-tuned models on our GPUS. Due to the security requirements of the financial domain, none of the clients' data could be sent to third parties, so it was necessary to store all data in our own environment.
I designed KAIA's agentic architecture and built the AI-based orchestration and logic. This involved fine-tuning and using RLHF to align KAIA's answers with those in the banking domain. I evaluated the accuracy and designed the ingestion pipelines to fetch weekly data from government websites. I also integrated the entire KAIA with the front end, managed the front end and DevOps teams, and researched and provided appropriate APIs and deployment guidelines to the front end and DevOps teams.
Automatic Video Ads Research and Generation
Built an end-to-end system in which AI agents own the full ad lifecycle, including market and competitor research, concept generation, creative production, and performance testing, with no human in the loop.
A reinforcement-learning-fine-tuned LLM serves as the self-verifying harness, continuously evaluating every ad against performance data. It automatically promotes the top-performing scalers, thereby eliminating the need for manual creative testing. I designed the reusable skills, context, and MCP tools that enable the agent fleet to improve each cycle with less intervention, and engineered a containerized pipeline that generates complete ad videos from just a concept and a product description.
OUTCOME
Output rose from 5 to 50 ads per platform per week, with visual consistency that exceeded leading commercial generation tools.
A reinforcement-learning-fine-tuned LLM serves as the self-verifying harness, continuously evaluating every ad against performance data. It automatically promotes the top-performing scalers, thereby eliminating the need for manual creative testing. I designed the reusable skills, context, and MCP tools that enable the agent fleet to improve each cycle with less intervention, and engineered a containerized pipeline that generates complete ad videos from just a concept and a product description.
OUTCOME
Output rose from 5 to 50 ads per platform per week, with visual consistency that exceeded leading commercial generation tools.
KAIA, My Compliance Expert
https://www.360factors.com/products/kaia-ai-compliance-expert/A React- and Python-based agentic enterprise application designed for specialized compliance workflows. The application uses a microservices architecture with 8+ services operating simultaneously, allowing individual services to continue running independently if another service goes down. The architecture was designed with scalability in mind. Within one year, the application was adopted by 8+ mid-sized US banks, each with a valuation exceeding $3 billion.
Education
2024 - 2025
Master's Degree in Computer Science
Georgia Institute of Technology - Atlanta, USA
2016 - 2020
Bachelor's Degree in Electrical Engineering
National University of Sciences and Technology - Islamabad, Pakistan
Certifications
AUGUST 2025 - PRESENT
AWS Cloud Certification
AWS
JUNE 2019 - PRESENT
Tableau
Udemy
JUNE 2019 - PRESENT
Deep Learning Specialization
Coursera
APRIL 2019 - PRESENT
Machine Learning
Coursera
Skills
Libraries/APIs
PyTorch, XGBoost, Claude API, WhatsApp API, OpenAI API, TensorFlow, Node.js, React
Tools
Apache Airflow, Azure OpenAI Service, Microsoft Copilot, Claude Code, Retool, Codex, Git, Tableau
Languages
Python, SQL, SPARQL, TypeScript, JavaScript, Snowflake, R, Bash
Frameworks
Spark, Agentic Frameworks, LangGraph, Next.js
Paradigms
Model Context Protocol (MCP), Testing, ETL
Platforms
Amazon Web Services (AWS), Docker, Google Cloud Platform (GCP), Microsoft Copilot Studio, CrewAI, Azure, Databricks, Kubernetes, AWS IoT
Storage
Data Pipelines, PostgreSQL, Datadog, Neo4j, Redis, Graph Databases, On-premise, PostgreSQL 10, Azure Cloud Services
Industry Expertise
Marketing
Other
Large Language Models (LLMs), Machine Learning Operations (MLOps), Machine Learning, Full-stack, AI Engineering, Agentic AI, Forecasting, Data Engineering, Demand Forecasting, Data Science, Time Series Forecasting, Model Evaluation, Communication, Model Development, Feature Engineering, Monitoring, Stakeholder Management, Model Deployment, Linear Regression, Artificial Intelligence (AI), Generative Artificial Intelligence (GenAI), Model Tuning, AI Modeling, Attention to Detail, Screeners, Written Communication, Code Review, Interviewing, API Integration, Agentic RAG Systems, Minimum Viable Product (MVP), OpenAI, Booking Systems, Localization, Mobile Apps, User Experience (UX), User Interface (UI), Stripe Payments, Optical Character Recognition (OCR), Data Analysis, Conversational AI, FastAPI, Solution Architecture, AI Agents, Multi-agent Systems, Answer Engine Optimization (AEO), Gemini API, Prompt Engineering, ChatGPT API, Technical Architecture, System Architecture, AI Integration, Fractional CTO, SaaS, System Architecture Design, Software System Architecture Development, Technical Leadership, Scalability, Software Engineering, Agentic AI Systems, LangChain, API Design, Performance, Performance Optimization, Vector Search, Microsoft Azure, APIs, Amplitude, AI Architecture, Context Engineering, RAG Systems, Ontologies, Temporal, Knowledge Graphs, Data Modeling, Business Analysis, Semantics, Data Build Tool (dbt), Technology, Strategy, Natural Language Processing (NLP), Security Audits, Data Security, Startups, SOC 2, Document Processing, Legal Technology (Legaltech), Hugging Face, Gemini, Google ADK, RAG Architecture, Cloud Platforms, LLM Fine-tuning, Statistical Forecasting, Fine-tuning, Team Leadership, Team Management, Supabase, SEO Tools, Audits, Auditing, CTO, Cloud Architecture, Amazon Bedrock AgentCore, Containerization, CI/CD Pipelines, Reinforcement Learning from Human Feedback (RLHF), RAG Pipelines, Bayesian Networks, Deep Learning, Reinforcement Learning, Software Development, Statistical Modeling, Computer Vision, Linear Control Systems, ARM Embedded, Cloud, Transformers, Embedded Systems
Collaboration That Works
How to Work with Toptal
Toptal matches you directly with global industry experts from our network in hours—not weeks or months.
1
Share your needs
Discuss your requirements and refine your scope in a call with a Toptal domain expert.
2
Choose your talent
Get a short list of expertly matched talent within 24 hours to review, interview, and choose from.
3
Start your risk-free talent trial
Work with your chosen talent on a trial basis for up to two weeks. Pay only if you decide to hire them.
Top talent is in high demand.
Start hiring