SYSTEMAMMAR.AI
00:00:00 PKT

► Field notes from a senior ML engineer

Ammar Waqar.

I design and ship production AI systems ; LLMs, computer vision, RAG pipelines, and multi-agent workflows that solve real-world problems at scale.

B.Sc. Artificial Intelligence
PK · Remote · Freelance
Senior ML Engineer · Nysonian
ammarwaqar.work@gmail.com →↓ CV+92 304 544 8920
MEM87%MODELS10 ONLINEUPTIME14d 6h 21mNODEPKT-01SIGNALGREENTEMP41°CLATENCY12msPKTS0 DROP
MEM87%MODELS10 ONLINEUPTIME14d 6h 21mNODEPKT-01SIGNALGREENTEMP41°CLATENCY12msPKTS0 DROP
MEM87%MODELS10 ONLINEUPTIME14d 6h 21mNODEPKT-01SIGNALGREENTEMP41°CLATENCY12msPKTS0 DROP
file_02

Ship log.

  1. Sep 2026 — Present

    Senior Machine Learning Engineer @ Nysonian

    • 01Built agentic AI systems that orchestrate customer-support workflows through tool use.
    • 02Built models for varied use cases, handling the workflow from data collection and cleaning through training and iteration.
    • 03Developed and maintained customer-support agents and n8n workflows to automate support operations.
    • 04Improved retrieval and prompt reliability to help agents deliver accurate, context-aware responses.
    • 05Integrated tool calling with databases, Shopify, and ShipHero so agents can retrieve operational data and respond to customer inquiries.
  2. 2025 — 2026
    Islamabad, PK

    Artificial Intelligence Engineer @ Intercraft

    • 01Developed AI-driven solutions for multiple national and international client projects across industries.
    • 02Contributed to AI system development for government-sector initiatives, including large-scale projects in Saudi Arabia.
    • 03Designed and integrated machine learning workflows for real-world applications and production systems.
  3. 2025
    Remote

    Associate Software Engineer (LLM Trainer) @ Sirovista / Turing

    • 01Evaluated LLM outputs as a human-in-the-loop judge to improve response accuracy and reliability across tasks.
    • 02Reviewed reasoning quality, factual correctness, and instruction adherence to guide iterative model improvements.
    • 03Designed prompt evaluation scenarios to identify model weaknesses and reduce hallucinations.
  4. Aug 2024
    NUST, Islamabad

    AI & ML Intern @ Arcelik R&D Center

    • 01Fine-tuned pre-trained LLMs on domain-specific datasets, improving downstream task F1-scores by 15% across 3 R&D projects.
    • 02Developed custom prompt templates and model variations, enhancing contextual accuracy in language generation tasks.
    • 03Explored advanced fine-tuning strategies (LoRA, PEFT) under senior guidance, improving efficiency and real-world applicability.
    • 04Documented findings and contributed to internal knowledge base, supporting future interns and researchers.
  5. Jan 2022 — Present
    Remote

    Freelance AI/ML & Data Engineering Consultant @ Fiverr & Upwork

    • 01Delivered 15+ AI/ML and data engineering projects for international clients, maintaining a 4.9/5 satisfaction rating.
    • 02Developed end-to-end ML pipelines (preprocessing, training, deployment) using TensorFlow, PyTorch, and Scikit-learn.
    • 03Built AI automation workflows with n8n and GPT models, streamlining repetitive client tasks and reducing manual workload.
    • 04Designed Power BI dashboards and data models, enabling stakeholders to derive actionable business insights.
file_03

Selected builds.

01 / 09Multi-agent · Full-stack generation

Autonomous AI Coding Agent.

Autonomous multi-agent system that generates complete full-stack applications (ERP, CRM, dashboards) from a single natural language prompt.

  • 01Designed a LangGraph workflow with specialized agents for requirement analysis, code generation, QA validation, and Docker-based deployment.
  • 02Implemented automated debugging and checkpoint-based version control using Docker log monitoring and conversational memory for iterative refinement.
PythonLangGraphLangChainDockerClaude
02 / 09Call QA · Bilingual NLP

Flint (Call QA).

AI-powered call quality platform that transcribes Urdu + English sales recordings, scores agents against a configurable rubric, and produces bilingual coaching reports.

  • 01Built an end-to-end async pipeline (React → FastAPI → Celery) that stores audio in MinIO, transcribes with Deepgram speaker diarization, and evaluates agents via a pluggable LLM layer (Ollama / OpenAI / Anthropic).
  • 02Implemented LLM-based QA scoring (0–100) across tone, guidance, helpfulness, and behaviour, plus bilingual summaries (Nastaliq + Roman Urdu) and prioritized coaching for the next call.
ReactFastAPICeleryDeepgramMinIOLLM
03 / 09OCR · Intelligent document processing

Planet OCR.

Full-stack multi-engine OCR / IDP platform that turns PDFs and images into validated, business-ready data through intelligent routing, extraction, and human-in-the-loop review.

  • 01Designed a LangGraph-style multi-stage IDP workflow (normalize, classify, map, validate, route) with specialized OCR engines and LLM-assisted field extraction across 16 document types.
  • 02Built REST API, signed webhooks, Xero/QuickBooks connectors, and a Next.js review dashboard — deployed with Docker, Celery workers, and quota/billing controls.
PythonFastAPICeleryNext.jsTypeScriptPostgreSQLRedisDockerKongLLM
04 / 09Real-time computer vision

Dynamic Vision.

Modular real-time computer vision system for analyzing CCTV and video streams across multiple domains.

  • 01Implemented detection modules: suspicious activity monitoring, construction worker safety compliance, and sports motion analysis.
  • 02Designed a decoupled architecture enabling independent deployment and extension of vision modules for different real-world use cases.
PythonOpenCVCCTVModular CV
05 / 09AI automation pipeline

Automated Book Generation.

End-to-end AI automation pipeline to generate books (outline → chapters → final compilation) using n8n and GPT-4.

  • 01Automated over 90% of the writing workflow, reducing manual effort and enabling multiple book generations with consistent quality.
  • 02Integrated Supabase for real-time data storage and workflow tracking, ensuring reliability and scalability.
n8nSupabaseGPT-4
06 / 09Voice · RAG · Customer service

Multi-Domain Support Bot.

Multi-domain voice-enabled chatbot for customer service, adaptable to any business via domain-specific knowledge bases.

  • 01Integrated ChatGPT-4 with LangChain for RAG-driven responses and Twilio for real-time voice calls, enabling natural human-like conversations.
  • 02Handled 75% of customer queries autonomously through RAG-based knowledge retrieval with fallback escalation to human agents.
GPT-4LangChainTwilioRAG
07 / 09Deep learning · Robotics

Robot Localization DL.

Deep learning model solving the robot kidnapping problem — predicting robot coordinates from LiDAR data and map inputs.

  • 01Achieved 93.5% accuracy within 20px and 97.4% within 25px on custom datasets generated through simulation.
  • 02Created and preprocessed large datasets, trained multiple architectures, and optimized with early stopping and learning rate scheduling.
PythonTensorFlowOpenCVLiDAR
08 / 09Sports analytics · CV

UFC Fight Analysis.

Computer vision system for UFC fight analysis — fighter detection, action logging, and performance-optimized video processing.

  • 01Achieved 90% accuracy in fighter detection using YOLO, enabling reliable tracking in recorded matches.
  • 02Automated logging of 500+ fighter actions per match through MediaPipe and OpenPose integration.
  • 03Optimized performance with CUDA/cuDNN, reducing processing lag by 50% compared to baseline runs.
YOLOMediaPipeOpenPoseCUDA
09 / 09Workflow orchestration · Distributed systems

Orchestrated Recurring Coordination.

Enterprise workflow orchestration platform for managing recurring tasks and automated coordination across distributed systems.

  • 01Built a distributed scheduler for recurring workflows with dependency-aware execution, retries, and failure isolation across worker nodes.
  • 02Designed operator dashboards for run status, SLA tracking, and automated coordination so teams can monitor and recover jobs without manual intervention.
TypeScriptNext.jsNode.jsRedisPostgreSQLDocker
10 / 10

Your project?

Bring me in early — problem framing, model design, deploy.

Start a thread →
file_04

Stack matrix.

Programming

04
  • · Python
  • · C/C++
  • · Java
  • · SQL

Frameworks

06
  • · TensorFlow
  • · PyTorch
  • · Keras
  • · Scikit-learn
  • · FastAPI
  • · Flask

AI / ML

07
  • · Supervised & Unsupervised Learning
  • · CNNs
  • · RNNs
  • · Transformers
  • · Model Optimization
  • · Classification
  • · Regression

Reinforcement Learning

04
  • · Q-Learning
  • · DQN
  • · A3C
  • · Multi-Agent RL

Computer Vision

04
  • · OpenCV
  • · YOLO
  • · Image Processing
  • · Object Detection & Tracking

Automation

08
  • · n8n
  • · Supabase
  • · Workflow Automation
  • · Agentic AI
  • · Tool Calling
  • · LLM Pipelines
  • · LangChain
  • · LangGraph

Data Engineering

04
  • · Data Preprocessing
  • · Feature Engineering
  • · Power BI
  • · RAG

Edge

05
  • · HTML
  • · CSS
  • · JavaScript
  • · DSA
  • · LeetCode (50+)

05 / education & milestones

B.Sc. Artificial Intelligence
University of Wah, Wah Cantt

CGPA 3.52

  • 01Presented research on deep learning–based robot localization (research paper under review).
  • 02Completed internships in AI/ML at Arcelik R&D Center and Codex Cue Software Solution.
  • 03Earned certifications in Deep Learning and AI from university-led and external programs.
  • 04Active participant in AI workshops, hackathons, and coding competitions at UoW and NUML.
0+
AI/ML projects delivered
0%
localization accuracy
0%
queries handled autonomously
0+
LeetCode problems solved

► End of file — start a new one

Let's train
something real.

ammarwaqar.work@gmail.comCall +92 304 544 8920