Emre mrkckpln

Software Engineer & AI Systems Architect

Architecting resilient, high-throughput distributed backends, parameter-efficient LLM pipelines, and production-grade intelligent SaaS platforms.

Rotakur: Unified Mobility & Fleet SaaS
Enterprise B2B Fleet Management + B2C Mobility Marketplace
A resilient, multi-tenant hybrid system orchestrating enterprise fleet operations, corporate logistics, and public mobility bookings.
- AI Dispatch & Spatial Routing: Engineered an NLP Smart Dispatcher (Gemini Flash + pgvector) to extract structured reservation profiles from raw chat text. Coupled with OSRM multi-stop road-network polyline decoding, dynamic geofencing, and strict POI validation to eliminate phantom bookings.
- Dual-Tenant Governance & P&L Engine: Governs multi-entity corporate procurement via a locked state machine (DRAFT -> APPROVED) with strict data isolation. Integrated an AI Surge Pricing engine calculating real-time binding fares and exact Net P&L margins prior to trip onset.
- Regulatory Compliance & Telemetry: Built a smart carpooling engine pairing B2C orders (±2h window / 20km radius) tracking DEFRA-compliant CO2 savings. Automates U-ETDS transport ministry reporting and executes UBL-TR 2.1 e-invoicing with DLQ-safe fallbacks.
Stack: Python, FastAPI, SQLAlchemy 2.0, PostgreSQL, pgvector, WebSockets, Redis, OSRM, Docker

♟️AI Chess Intelligence & Analytics
Proprietary Hybrid-Engine B2C SaaS & OTB Tournament Platform
High-performance analytics ecosystem bridging brute-force engine computation with abstract human positional and psychological understanding.
- Two-Tier Hybrid Engine: Orchestrated an ultra-low-cost Tier-1 client exploration layer (Stockfish 16 NNUE Wasm, 0 DAK cost) and decoupled Tier-2 deep server analytics (RQ worker pool, depth 20+) protected by transactional credit (DAK) quotas.
- OTB Intelligence & Scouting: Built custom telemetry modules parsing %clk zeitnot time-trouble, round-by-round cognitive fatigue curves, and automated FIDE profile scraping for pre-match tactical dossiers.
- Resilient Data Integrity: Enforced composite uniqueness (uploaded_by_id, checksum) for collision-free multi-user archives and atomic SQL-level XP progression pipelines with zero Alembic schema drift.
Stack: FastAPI, Next.js 15, PostgreSQL, Redis Queue (RQ), Stockfish 16 NNUE, WebAssembly, GPT-4o, fpdf2

👁️ Autonomous AI Sentry (Long-term Internship Project)
Real-Time Anomaly & Spatial Variance Perception
An autonomous perception engine designed for millisecond-latency behavioral monitoring and spatial safety tracking in mission-critical environments.
- Perception Architecture: Deployed YOLO-OBB and Pose architectures to infer anomalous trajectories, falls, and perimeter breaches with high temporal consistency.
- System Optimization: Streamlined computational graphs and data pipelines on Arch/CachyOS to minimize memory footprint and maximize frame throughput.
Stack: Python, PyTorch, OpenCV, Linux (CachyOS), FastAPI

Distributed Systems & Data-Intensive Backends: Designing resilient, concurrent services adhering to strict Router → Service → Repository patterns, ACID isolation, transactional state machines, and asynchronous worker queues (Redis/RQ).
First-Principles Deep Learning & Mathematics: Deconstructing computational graphs, reverse-mode autodiff engines, loss gradients (MSE, Cross-Entropy), adaptive optimizers (SGD to AdamW), and Transformer Self-Attention mechanics from scratch.
Model Adaptation, PEFT & Alignment: Modifying pretrained foundation architectures via Parameter-Efficient Fine-Tuning (LoRA, QLoRA 4-bit NF4, paged optimizers), instruction tuning (Alpaca/ShareGPT JSONL formatting), DPO alignment, and model compression (AWQ/GGUF).
AI Engineering & High-Throughput Inference: Architecting low-latency serving pipelines leveraging KV-caching, continuous batching (vLLM PagedAttention), Advanced Hybrid RAG (Dense HNSW + Sparse BM25 with Cross-Encoder reranking), and deterministic ReAct agentic workflows.

Systems Rigor (DDIA & Concurrency): Systematically studying distributed consensus, replication strategies, and fault-tolerant data pipelines via Martin Kleppmann's DDIA and low-level concurrency models.
Strategic System Foresight: Translating 15+ years of licensed competitive chess expertise into engineering — proactively mitigating race conditions, architectural drift, and system bottlenecks before they manifest.
First-Principles Mathematical Formulation: Practicing the Feynman technique to deconstruct complex ML architectures, loss surfaces, and vector calculus into executable codebases from scratch.
Multilingual Global Engineering: Operating with professional fluency in English and actively acquiring German (B-level) for cross-border technical communication.