Developed a 2D space shooter with advanced enemy AI and
power-up systems. Features include particle effects, sound
design, and progressive difficulty scaling.
Unity
C#
Created an enhanced clone featuring multiple platform types,
power-ups, and a competitive leaderboard system with cloud
saves.
Unity
C#
Match-3 word game inspired by AlphaBetty with an Alan
Wake-inspired theme. Implemented core word-matching mechanics,
gameplay systems, and dynamic interactions.
Unity
C#
Unity Tower Defense game built to collect player item data and
analyze gameplay behavior. Integrated a Django backend with
LLMs to generate automated item-balancing recommendations.
Unity
Django
Python
Built an AI sales agent that introduces products, answers
customers, negotiates prices, and guides conversations toward
purchases. Added product tools, pricing logic, and proactive
follow-ups to automate multi-step sales interactions.
Python
LLMs
Tool Calling
Telegram
Voice-based AI call center combining ASR, RAG, LLMs, and TTS
for real-time conversations. Integrated speech recognition,
knowledge retrieval, response generation, and speech
synthesis into a low-latency pipeline.
ASR
RAG
LLMs
TTS
Bilingual (English/Persian) AI-powered fitness application.
Flutter frontend with a FastAPI backend built on PostgreSQL,
Redis, Docker Compose, Nginx, JWT auth, and Google OAuth.
Designed around AI-assisted fitness functionality and
personalized interactions — an idea turned into an
end-to-end product.
Flutter
FastAPI
PostgreSQL
Redis
Docker
JWT
Systems where LLMs reason, select tools, and call functions
against APIs, databases, and services to complete multi-step
workflows. Focus on structured outputs, context management,
orchestration, and error handling — running on both cloud and
local models.
Python
LLMs
Tool Calling
FastAPI
Multi-stage Persian ASR pipeline combining Whisper /
Faster-Whisper with Silero VAD, noise reduction, and speech
enhancement (DeepFilterNet, Demucs), plus LLM-based
post-processing for punctuation restoration and Persian text
normalization to improve transcription quality.
Python
Faster-Whisper
Silero VAD
FFmpeg
Self-hosted model serving on Linux + NVIDIA GPU servers using
llama.cpp, Ollama, and vLLM with GGUF quantized models.
Hands-on work with context-window optimization, KV cache /
VRAM tuning, systemd services, journalctl, and OpenAI-style
model APIs behind Nginx.
llama.cpp
Ollama
vLLM
CUDA
Linux