Malik I.

I build AI systems that work in production — not just demos. With 5+ years of professional engineering experience across AI systems, defense R&D, and software products with 500,000+ end-user downloads, I deliver architected solutions — not experiments. I am one of the few engineers who combines deep LLM/multi-agent orchestration expertise with real Unity simulation and game integration experience. This means I can take on AI projects that require both intelligent decision-making and high-performance real-time environments. ━━━ WHAT I BUILD ━━━ → Multi-Agent AI Systems (LangGraph, CrewAI) — autonomous pipelines with tool-use, memory, reflection loops, and human-in-the-loop controls → Model Training & Fine-Tuning — custom domain adaptation for LLMs and vision models using parameter-efficient fine-tuning (PEFT, LoRA). → RAG Pipelines — Hybrid Search (BM25 + vector), Reciprocal Rank Fusion, ChromaDB, FAISS, Pinecone → LLM Integration — Claude API, GPT-4o, DeepSeek V3, Groq/Llama 3, prompt engineering, structured outputs → AI for Unity — ML-Agents (PPO/SAC), intelligent NPCs, computer vision pipelines (Unity Sentis/ONNX), RL-based agents → Production Backends — FastAPI Python servers, async APIs, Celery + Redis task queues, and containerized deployments (Docker). ━━━ PROVEN SYSTEMS ━━━ ▶ Foreman (AI Software Factory) — Production-grade multi-agent orchestration system. Brain + Worker architecture, 5-level self-healing, loop detection, HybridRetriever, secure file locking, multi-provider LLM support (DeepSeek, GPT-4o, Claude). ▶ Scrivio (Academic AI Agent) — 7-agent LangGraph pipeline that autonomously writes full academic research papers with verified DOIs and zero hallucinated citations. Stack: FastAPI · Groq · DeepSeek · Celery · Redis · ArXiv · PubMed · Semantic Scholar APIs. 89 backend + 9 E2E tests all passing. ▶ AWC Vision Systems — AI/ML computer vision pipelines integrated directly into Unity runtime for real-time object detection and tracking, built for one of Pakistan's most sensitive defense R&D organizations. ▶ Faris Drone Simulation (POF) — High-fidelity drone flight simulation in Unity for Pakistan Ordnance Factories, modeling real-world aerial dynamics, physics, and flight-path algorithms for defense evaluation. ▶ BioScout — Hybrid RAG agent (BM25 + ChromaDB + DeepSeek Vision) for hallucination-free domain identification and Q&A, built under hackathon conditions. ▶ Bias Detection System (XAI) — Explainable AI pipeline using SHAP analysis to detect sentiment, gender, and toxicity bias in LLM outputs. Auditable feature-level attribution for enterprise LLM safety and compliance. ▶ RL Robotic Arm — Reinforcement learning agent (PPO) trained in Unity ML-Agents to perform stable pick-and-place tasks in randomized environments. Integrated with ROS2 and a Niryo OS2 digital twin. ▶ PINN Research — Physics-Informed Neural Networks applied to the Heimburg nerve impulse soliton model. Embeds governing physical equations into neural network training for data-efficient scientific simulation. Paper in submission. ━━━ TECH STACK ━━━ LLMs & Agents: LangGraph · CrewAI · LangChain · Claude API · GPT-4o · DeepSeek · Groq · Llama 3 Training & Fine-Tuning: PyTorch · Hugging Face · LoRA / QLoRA · PEFT · Custom Model Training RAG & Search: ChromaDB · FAISS · Pinecone · rank_bm25 · Hybrid Search · RRF Backend & DevOps: FastAPI · Python · Celery · Redis · Docker · AWS/Cloud Deployment · REST APIs Unity & Simulation: Unity 3D/C# · ML-Agents · Unity Sentis (ONNX) · Reinforcement Learning · ShaderGraph Other: ROS2 · Digital Twin · SHAP/XAI · PINNs · Computer Vision ━━━ I SPECIALIZE IN ━━━ ✔ Agentic business automation — multi-agent workflows that operate 24/7 ✔ Domain-specific model adaptation — fine-tuning and deploying custom models for niche industries. ✔ Academic, legal & scientific AI — domain RAG pipelines, PINNs, research agents ✔ Explainable AI (XAI) — SHAP-based bias detection, safe and auditable LLM outputs ✔ AI + Unity integration — intelligent NPCs, ML-Agents, vision pipelines in game/sim environments All my projects are real, documented, and live on GitHub. I welcome technical interviews and architecture discussions before engagement. Open to hourly and fixed-price contracts. Let's build something intelligent.