whoami
Abdullah Ahmad
$ idea to production: RLHF, agentic LLM systems, data pipelines
AI & Data Engineer. Shipped RLHF pipelines, agentic LLM systems, applied ML models, and automation workflows for Microsoft, Moonshot AI, and Penguin AI. Three years taking AI products from zero to deployed.
- role
- AI / Data Engineer
- based
- Lahore, PK
- focus
- applied ML · agentic LLM · RLHF · automations
- status
- available for work
About
I take AI products from idea to production. Over 3 years I've shipped RLHF training pipelines, agentic LLM systems, applied ML models, and automation workflows for clients including Microsoft, Moonshot AI, and Penguin AI.
My work spans model quality and production engineering: adversarial evaluation, fine-tuning dataset curation, multi-agent orchestration, n8n and LangGraph automation workflows, and the data pipelines (Kafka, Airflow, BigQuery) that keep systems reliable at scale. On the consulting side I've guided 10+ companies from discovery through MVP delivery across fintech, SaaS, and analytics.
Based in Lahore, open to remote.
Experience
- Jun 2025 - PresentLahore, PK · Full-time
AI & Data Engineer
Analytiverse
- +Shipped RLHF-grade training data and agentic LLM systems for Microsoft, Moonshot AI (KIMI), and Penguin AI. 100+ preference datasets fed directly into client fine-tuning pipelines.
- +Designed 400+ adversarial prompts across 6 domains; built an LLM-as-a-Judge evaluation framework benchmarked against human annotations for grammar, factuality, and completeness.
- +Built multi-turn agentic workflows (LangGraph + LangChain) with sequential and parallel tool calls; multi-agent orchestration achieved 40%+ gains over single-agent baselines.
- +Migrating 4+ legacy pipelines to GCP using Kafka, Airflow, and BigQuery (medallion architecture); also building an AI cybersecurity log analyzer with ETL to BigQuery and a natural-language chatbot layer for anomaly detection.
- Oct 2025 - May 2026Remote · Contract
AI Product Consultant
Ember AI · RemoteBase
- +Consulted 10+ clients end-to-end across fintech, SaaS, and analytics: from discovery and PRD through MVP scoping to delivery, fully async across multiple concurrent engagements.
- +Shipped a natural-language data query chatbot over enterprise data warehouses using LangChain and RAG, replacing ad-hoc SQL requests for non-technical stakeholders.
- Apr 2025 - Jun 2025Lahore, PK
Generative AI Intern
Systems Limited
- +Built an agentic developer-assistant that analyzes code context and suggests improvements; integrated RAG over internal documentation, raising suggestion accuracy.
- +Designed automated bug-detection agents using AST parsing to flag syntax and logic errors, then optimized the response-streaming pipeline for lower perceived latency.
- Jan 2023 - PresentLahore, PK · Freelance
AI & Software Consultant
Independent
- +Built a full-stack inventory management system (Vite + Node.js + MongoDB) handling 15+ product categories and thousands of SKUs, with an AI-assisted reorder and demand-forecasting layer.
- +Delivered e-commerce builds and AI automation tooling for SME clients; managed projects end-to-end from scoping through deployment.
Technical skills
- Generative AI
- RAGLangChainLangGraphPrompt EngineeringSFT / RLHF DataLLM Evaluation
- Applied ML
- PyTorchTensorFlowTransformersComputer VisionNLPScikit-learn
- Automation & Workflows
- n8nWorkflow AutomationAPI IntegrationsAirflowKafka
- MLOps & Infrastructure
- DockerFastAPIGCPBigQueryGitLinux
- Languages
- PythonSQLC++JavaScript
Selected work
- 01private
AI Security Log Analyzer
data engineering · LLM · automationEnd-to-end cybersecurity pipeline: raw logs ingested via Kafka, transformed through a medallion architecture in Airflow, landed in BigQuery. A natural-language chatbot layer lets analysts query anomalies and threat patterns without writing SQL.
PythonKafkaAirflowBigQueryLangChainGCP - 02repo
Lexio
RAG · legalA retrieval pipeline for legal documents: DeepSeek-R1 for reasoning, FAISS for vector search, and prompt engineering that grounds every answer in retrieved context, so the model cites, it doesn't invent.
PythonLangChainFAISSDeepSeek-R1 - 03repo
Static Bug Analysis Agent
NLP · codeA BERT classifier fine-tuned on code vulnerabilities to rank bug severity, with the Gemini API layered on top for plain-language explanations and fix suggestions next to each finding.
PythonBERTGemini API - 04repo
Text Restore
vision · super-resImage restoration for degraded Urdu text. SCUNet handles denoising, SwinIR the super-resolution, trained on a synthetic dataset with controlled noise and blur, built for a low-resource script.
PythonPyTorchSwinIRSCUNet - 05private
Inventory Management System
full-stack · AI · automationFull-stack inventory system for a multi-category retail operation: 15+ product categories, thousands of SKUs, real-time stock tracking, and an AI-assisted demand forecasting layer that flags reorder points automatically.
ViteNode.jsMongoDBPython
cat education.md
Education
BSc, Computer Science
FAST-NUCES
Contact
Open to AI/ML engineering roles and collaboration. Email is fastest; I read everything.