Muhammad
Afaq Jamshaid
Applied AI & ML engineer shipping end-to-end GenAI, agentic, and computer-vision systems - and evaluating the LLMs behind them.
Software Engineer / AI Specialist
@ Mountain Creative
Germany / Remote
AWS AI Practitioner / OCI GenAI Pro
IturnbusinessproblemsintoproductionAI-GenAIapps,agenticsystems,MLpipelines,andcomputer-visionmodelsthatship.
$ whoami
Applied AI & ML engineer. I turn business problems into production AI systems - end to end, from data to deployment.
$ cat expertise.txt
- > GenAI & RAG applications
- > Agentic AI - Anthropic SDK, Agno (beyond chatbots)
- > LLM evaluation - RLHF / DPO / SFT
- > ML pipelines - supervised & reinforcement learning
- > Computer vision & Vision-Transformer research
$ status Open to AI/ML roles & collaborations
// tooling & stack
Skills
$ git log --author=afaq
Experience
Software Engineer - AI Specialist
Mountain Creative
Remote
- >Built an invoice-classification system: labeled 10,000+ invoices with a local LLM (Ollama), then trained a SetFit (transformer + logistic regression) classifier that sorts business invoices into 50+ categories.
- >Integrated a zero-shot object detection model (SAM3) via Roboflow into an AI-agent pipeline that automatically flags visual discrepancies in retail shelf images against expected conditions.
- >Developed and deployed web-based AI tools using Next.js, Supabase, and Vercel.
Research Assistant
BTU Cottbus-Senftenberg
Cottbus, Germany / Hybrid
- >Contribute to pandapipes, an open-source Python library for energy-pipeline simulation, resolving bugs during feature migration from pandapower.
- >Building a local LLM-based diagnostics layer that grounds explanations in stack traces, user code, and source context to generate actionable error analysis for simulation workflows.
- >Developed a Plotly + Mapbox module to visualize pipeline networks on real-world maps, and authored technical docs & tutorials for onboarding.
Applied AI Engineer - Intern
Lund Combustion Engineering (LOGE AB)
Cottbus, Germany / On-site
- >Replaced a manual search process with an ML model predicting valid fuel combinations within a 25x25-dimensional space of millions of possibilities.
- >Delivered a customized end-to-end ML pipeline - dataset creation, training, prediction, visualization - achieving 93% prediction accuracy.
- >Designed a hybrid RL training strategy: online learning for feature-importance feedback + offline learning via iterative data injection.
LLM Evaluation Engineer
Freelance
Remote
- >Benchmarked and evaluated LLMs including ChatGPT, Claude, and LLaMA against scenario-based unittest and pytest cases, assessing compliance with the requirements rubric.
- >Built supervised fine-tuning datasets and scripts across Python, C++, and JavaScript, generating 1000+ structured prompts and reference responses for SFT and RLHF workflows.
// selected work
Projects
SMENTRY
Access-control system using computer vision to detect and authorize resident number plates - 25% faster entry processing.
- >Trained a custom YOLOv8 model + OCR, cutting manual entry errors by 30%.
- >Integrated the model into the web portal, reducing processing time by 20%.
MyCFO
AI financial assistant that parses natural-language expenses and generates real-time budget insights via Gemini + Groq.
- >Structured expense parsing & categorization with Gemini for consistent ledgers.
- >Responsive Tailwind dashboards for budget trends and spending insights.
HireScore
LLM-driven CV-JD matching with LangChain + Ollama: analyzes alignment, detects skill gaps, auto-generates JD-based quizzes.
- >Chroma for semantic storage & retrieval of JDs and candidate profiles.
- >FastAPI + SQLite backend for quiz results, scoring, and the React front end.
Hybrid AI Fuel Pipeline
Supervised + reinforcement learning pipeline to optimize complex fuel blends across multiple targets, cutting simulation cost.
- >Feature-weighted scoring + zoom-in data generation to avoid local minima.
- >Continuous supervised<->RL improvement loop for faster optimization cycles.
ClarityQA
Streamlit RAG chat app with document upload, vector search, and context-aware answers powered by Mistral LLM + Chroma.
- >Accurate, grounded Q&A over uploaded documents via RAG.
- >Persistent vector storage with efficient session and resource management.
$ verify --credentials
Certifications
// let's connect
Building something with AI?
Open to AI/ML roles, research collaborations, or a chat about LLMs, agents, and shipping models to production.