I build AI that thinks, plans, and adapts — not just predicts.
🚀 Hey, I'm Ammar
I work at the intersection of large language models, cognitive science, and systems engineering. My focus is on building reasoning-first agents and modular multi-agent frameworks that go beyond prompt-response.
- 🧠 Core interest: agents with belief modeling, memory, and long-term goals
- 🔍 I care about making AI useful, interpretable, and actually smart — not just "chatty"
🧠 What I Work On
- Cognitive agents — agents that track goals, beliefs, and mental states
- Reasoning pipelines — LLM stacks designed for logic, not just fluency
- Memory systems — long-term memory, self-reflection, and behavior updates
- Tool-using agents — LangGraph, OpenAI Assistants, Google AI SDK, LangChain
- Multi-agent systems — coordination, planning, modular workflows
🛠️ Projects
- Zirel-0 – DeepSeek-R1-Zero finetuned for better reasoning and minimal hallucination
- OR1 Behemoth – Qwen 3 72B trained for Rust — strongest Rust completion & Generation model available
- CogniLink – LLaMA 3.3 70B with added reasoning + reflection, without losing creative edge
- Manticore 32B – Reasoning-tuned Qwen3-32B with low-latency high-accuracy outputs
- VisoLearn – Multimodal VLLM tutor for neurodivergent learners (ASD)
- Qwen Griffon – Lightweight SLM with enhanced reasoning at low cost
- Self-Learning RAG – Retrieval system that adapts to user context and evolves
- Enlightener – RLHF-trained tutor agent with long-memory and personalized feedback
- Immy – Emotion-aware real-time AI bear with adaptive state modeling
- Wisal - a revolutionary, AI-powered educational platform designed specifically for children with Autism Spectrum Disorder (ASD) in real time.
🧰 Tools I Use a Lot
Languages: Python, Rust, C++, Java
Frameworks: PyTorch, Transformers, TRL, Unsloth, Accelerate , Burn , Lightning
Agent Stacks: LangGraph, LangChain, OpenAI Agents framework, Google ADK (Google AI Agents) ,Pydantic AI
Focus Areas: RAG, memory-augmented agents, symbolic logic, multi-agent systems , Small language models (Small VLMs too)
📡 Let’s Connect
Building systems that learn, reason, and adapt. No fluff. Just logic.