I build and study language models, inference systems, agentic software, and applied AI.
research idea → experiment → benchmark → runtime → product
Language-model systems that are smaller, faster, measurable, and actually useful.
- Native inference runtimes and portable model formats
- Alternative LM architectures, routing, and tokenization
- KV-cache efficiency, quantization, CUDA/Triton kernels
- Turkish SLMs and evidence-grounded reasoning models
- Tool-using agents and local-first AI products
| Project | Focus |
|---|---|
| RelationLM | Experimental language-model architecture built around explicit anchor–partner relations, sparse routing, RelationLex tokenization, and custom Triton selection kernels |
| ForesightLM | Sentence-level future-signal learning, calibration, semantic reranking, and self-loop hard negatives in small language models |
| NedoLM-pt | Training, SFT, tokenizer tooling, and reproducibility code for the NEDO Turkish SLM project |
| RiskKV | Risk-aware heterogeneous KV-cache compression for preserving exact-critical context under aggressive compression |
| QA-Research-SLM | Evidence-grounded research reasoning, claim verification, structured outputs, and verifier-guided small-model training |
Foresight Is Not Enough: Sentence-Level Future Signals, Self-Loop Hard Negatives, and the Calibration Gap in Small Language Models
| Project | Focus |
|---|---|
| MercanRuntime | Native local inference stack for Mercan/NedoLM models with a portable .mercan format, C ABI, Architecture/Tokenizer SDKs, and experimental Graph ABI |
| SLMOrchestration | Local workspace for streaming Ollama/Hugging Face inference, runtime controls, model presets, and attachments |
| RelationLM | Stateful sparse decode experiments and specialized Triton kernels for relation-based inference |
| RiskKV | Memory-efficient long-context inference through heterogeneous KV fidelity |
C · C++ · Rust · CMake · Linux · CUDA · Triton · Docker
| Project | Focus |
|---|---|
| BrowserAgent / Ethgent | Multi-agent orchestration for browser automation, research, content creation, and tool-driven workflows |
| MarketingStudio | Ad creatives, short-form video pipelines, local image generation, and social connectors |
| CNCapp / SacFlow AI | AI-assisted DXF processing, nesting, manufacturing analysis, pricing, and quotation workflows |
| DataMEDX2 | Oncology analytics, clinical decision-support experiments, and AI-assisted data processing |
| TUA | Lunar-rover path planning and autonomous-navigation experiments |
| VoltranScout | FTC scouting and analytics tooling |
AI / ML
Transformers · PEFT / QLoRA · Triton · LLM/SLM training · quantization · tokenization
Systems
C · C++ · Rust · CMake · Linux · native inference
Product / Backend
Node.js · Vite · SQLite · Redis · FFmpeg