I take AIintoproduction.

Muaaz Ahmad

Forward Deployed AI Engineer · Riyadh, Saudi Arabia

Five years embedding with business teams to turn their real workflows into agentic systems, RAG and document AI that run in production.

Download CV Email me

Embedded with the team. Accountable for production.

I work where the problem lives. I sit with the people who own the workflow, find what actually slows them down, and build the system that changes it, from the first prototype to the production rollout and the handover.

Most of that work is in documents that can’t leak or lie: legal, healthcare and finance. So the systems cite their sources, gate releases on evaluations, redact PII and keep tenants isolated.

8 → 25+fps: Real-time sports video analytics, after custom YOLO training and TensorRT quantization.
10×faster: Inference latency, from async FastAPI endpoints, request batching and Redis caching.
150k+installs: Across two consumer GenAI apps: Pictro AI (100,000+) and ToonMe (50,000+).
4engineers: Led as AI Team Lead, with code review, testing standards and deployment workflows.

Systems I’ve shipped

Client and employer work, plus open-source builds you can read line by line. Each diagram is drawn from the system itself.

LegalMind

Hybrid retrieval, mandatory citation validation and RAG Triad thresholds in CI. Deployed on GKE with HPA autoscaling.

LegalMind
Legal knowledge assistant · open source
Production-grade RAG for legal Q&A that refuses to answer without a source: every claim is validated against the text, then challenged by three evaluation agents.
Python · FastAPI · Llama 2 · PEFT · LangGraph · ChromaDB · Pytest · GKE
Read the code
Hybrid retrieval, mandatory citation validation and RAG Triad thresholds in CI. Deployed on GKE with HPA autoscaling.
DocMind
Multi-tenant RAG platform · open source
Upload documents, ask questions, get streamed answers with citations pointing at the exact source chunk. Every tenant is isolated down to the vector collection.
Next.js · FastAPI · PostgreSQL · Redis · ChromaDB · LangChain · Cohere · Docker
Read the code
Next.js 15 and FastAPI over PostgreSQL, Redis and ChromaDB. Multi-tenant isolation, on GKE with Prometheus monitoring.
MineOps Copilot
Operations copilot for mining · in progress
A bilingual copilot that answers SOP questions with citations, queries operations data through governed MCP tools, and drafts work orders that wait for a human signature.
Python · LangGraph · OpenSearch · Qdrant · MCP · vLLM · NestJS · Kubernetes
Read the code
In progress, on synthetic data for a fictional mining company. FastAPI, vLLM/Ollama, NestJS, Helm/Kubernetes, OpenTelemetry.
No-Code Agent Builder
Rapid Labs · in production
A platform for non-technical users to create and deploy their own AI agents, running on AWS ECS with autoscaling and load balancing at sub-second latency.
LangChain · LangGraph · Qdrant · PostgreSQL · FastAPI · Docker · AWS ECS
Rapid Labs. A platform that lets non-technical teams create and ship their own AI agents, with sub-second latency.
Sports Video Analytics
Real-time tracking
Custom YOLO training to track players and the puck, then quantized with TensorRT until the pipeline ran in real time: from 8 to more than 25 frames a second.
PyTorch · YOLO · TensorRT · Gemini API · AWS GPU · Docker
PyTorch, YOLO, TensorRT, Gemini API on AWS GPU, in Docker.
Pictro AI & ToonMe
Consumer GenAI apps
An AI art and anime generator and an AI photo editor, with diffusion, GANs and face restoration behind them. Together, more than 150,000 downloads.
Stable Diffusion · SDXL · GANs · GFPGAN · ESRGAN · FastAPI · Cloud Run
An AI art and anime generator and an AI photo editor, served from Docker on GCP Cloud Run.

Also on GitHub

Where I’ve worked

Six teams since 2020, from Python services to production agent platforms.

2025 — now
Accelerated Innovation Group
Senior AI Engineer · Georgia, USA · remote
Architecture guidance for enterprise teams taking LLM systems to production: RAG, multi-agent orchestration, evaluation gates in CI.
2024 — 2025
Rapid Labs
Machine Learning Engineer · Islamabad
No-code multi-agent platform on AWS ECS, an LLM evaluation harness that blocks bad merges, 10× lower inference latency.
2023 — 2024
Metex Labz
AI Team Lead · Islamabad
LangGraph legal research agents with citation validation, Llama 2 fine-tuned on legal text, a team of four ML engineers.
2022 — 2023
MetaFourt
Team Lead, AI Solutions · Faisalabad
End-to-end NLP and computer vision delivery, containerised and tested in CI.
2021 — 2022
Hive Technologies
AI Engineer · Pakistan
Healthcare imaging models at the edge with TFLite; event-driven pipelines on Google Cloud and Vertex AI.
2020 — 2021
PhotonSofts
Python Developer · Faisalabad
Python web applications with AI-powered keyword analysis and search.

Toolbox

Agents & LLMs: LangGraph, LangChain, LlamaIndex, MCP, structured extraction, LoRA / QLoRA fine-tuning, GPT, Claude, Gemini, Llama, Mistral
Retrieval & evaluation: Hybrid search, Graph RAG, reranking, RAGAS, RAG Triad, faithfulness scoring, regression gates in CI
Document AI: AWS Textract, Azure Form Recognizer, Tesseract, contract and invoice parsing, table extraction
Build & ship: Python, FastAPI, Node.js, Next.js, PostgreSQL, Redis, vector DBs, Docker, Kubernetes, AWS, GCP, vLLM
Privacy & security: PII redaction, tenant isolation, encryption in transit and at rest, RBAC, secrets management