I am a **Software Engineer and Computer Science graduate ** specializing in LLM systems, autonomous agents, and resilient application architectures including both front end and backend designs.
My work sits at the intersection of deep learning internals, systems design, and real‑world products (mobile, desktop, and web).

I care less about “calling an API” and more about how models work under the hood, how they are orchestrated, and how they behave in production.

• LLM internals & math: Transformer forward pass, attention scoring, optimization tricks, KV caching, and efficient decoding.
• Hybrid architectures: Combining Transformers with State Space Models (Mamba/SSM) for long‑context and streaming workloads.
• Agentic orchestration: Planning/execution patterns, tool‑use, memory routing, and JSON‑schema‑driven supervisors.
• AI‑native backends: FastAPI + vector DBs + queues, designed around observability, latency budgets, and failure handling.

From‑scratch implementations of core deep learning algorithms and architectures to demystify how modern networks really work.
Highlights: Manual backprop, optimization routines, and neural architectures implemented without hiding behind high‑level frameworks.

Role: Research engineer mindset — building blocks, not just using them.
Stack: Python, NumPy, Jupyter

End‑to‑end fraud detection pipeline for imbalanced financial transaction data.

• Data preprocessing, feature engineering, and model training tuned for highly skewed labels.
• Evaluation focused on precision/recall trade‑offs and practical deployment considerations.

Stack: Python, scikit‑learn, Pandas, ML pipelines

Predictive modeling project for diabetes risk classification, with careful handling of class imbalance and interpretability.
Stack: Python, scikit‑learn, exploratory data analysis, classical ML models

A desktop‑first, cross‑platform planner that treats time and attention as first‑class resources.

• Energy‑aware scheduling that aligns tasks with cognitive peaks.
• Clean Architecture with repositories and strictly separated layers.
• Polished, glassmorphic UI built for daily use.

Stack: Flutter 3 · Dart · Drift ORM · SQLite · Clean Architecture

A native Android media player focused on latency, offline reliability, and delightful interactions.

• Advanced playback engine with PiP, gestures, and robust background playback.
• Reactive MVVM with StateFlow powering a modern Compose UI.

Stack: Kotlin · Jetpack Compose · Media3 · Room · Dagger Hilt

An offline‑first, zero‑knowledge diary with cryptographic guarantees.

• PBKDF2‑derived keys and encrypted storage for journal entries.
• Hardware‑backed keystore/keychain integration and biometric auth.
• Resilient autosave & recovery strategies tuned for real users.

Stack: Flutter · Dart · ObjectBox · Cryptography

Foundations: Neural nets, backprop, optimization, attention mechanisms
LLM / Agents: LangChain, custom tool-using agents, JSON-schema supervisors
Infra: Python, FastAPI, Qdrant, Redis, Ollama, REST/gRPC, background workers