AI Systems
LLM applications, gateways, agents, RAG, speech, NLP, multimodal workflows, and provider routing with explicit failure handling.
Building dependable software systems
Mohammad Hossein Rezayati
AI & Backend Engineer
I build reliable AI workflows, backend systems, automation platforms, product integrations, and production infrastructure.
AI workflowsrouting · agents
API layerFastAPI · WebSocket
StatePostgreSQL · Redis
OperationsDocker · CI/CD
Engineering capabilities
From applied AI to production operations
I work across system boundaries, connecting model capabilities, durable backend services, client experiences, and operating infrastructure.
Backend & Platform
FastAPI services, REST and WebSocket interfaces, PostgreSQL, Redis, queues, background jobs, and observable integrations.
Automation & Integrations
Messaging workflows, browser automation, retry and failover, orchestration, state reconciliation, and provider connectivity.
Product Engineering
TypeScript products, browser extensions, desktop integrations, dashboards, RTL interfaces, PDF workflows, and deployment.
Selected work
Systems built around real engineering constraints
Public projects spanning AI automation, service architecture, product integrations, computer vision, data, and applied machine learning.
AI & Automation
OCTOPUS
Durable Telegram control plane for AI-assisted software engineering with persistent sessions, browser automation, artifacts, and voice workflows.
- Python
- PostgreSQL
- Redis Streams
- Playwright
Backend & Platform
Ajil Unified AI Gateway
Multi-provider AI gateway with routing, failover, observability, WebSocket support, and multimodal interfaces.
- FastAPI
- Redis
- WebSocket
- LLM APIs
Product & Client
Lemon Editor Extension
TypeScript monorepo combining a Chrome MV3 editor extension, Express service, rich editing, PDF generation, and RTL/LTR support.
- TypeScript
- Manifest V3
- Express
AI & Automation
Speech-to-Text FastAPI Service
Speech pipeline with FFmpeg preprocessing, local and remote transcription providers, model management, jobs, dashboard, and OpenAPI.
- FastAPI
- Whisper
- FFmpeg
- Docker
Data & ML
Answer Sheet OMR
Configurable computer-vision pipeline for answer-sheet alignment, bubble detection, grading, PDF batches, and explainable artifacts.
- Python
- OpenCV
- Computer Vision
AI & Automation
EchoLingua
Multilingual TTS and audio-generation workflow with provider routing, recipe-based plans, caching, persistence, and Telegram delivery.
- Python
- Edge TTS
- FFmpeg
- Telegram
Data & ML
Triak Trade
Guarded AI-assisted signal research and backtesting platform with parsing, dashboards, and safety-focused execution boundaries.
- Python
- Backtesting
- Research
- Dashboard
Backend & Platform
Kiwicore
Reliable Telegram-to-Bale media relay with moderation, retries, idempotency, persistent state, and Dockerized services.
- Python
- Redis
- MySQL
- Docker
Backend & Platform
Gemini API Proxy Service
FastAPI and Cloudflare Worker gateway with multi-key rotation, retry and failover, access controls, monitoring, and load-test support.
- FastAPI
- Cloudflare Workers
- Failover
- Monitoring
AI & Automation
Agentic Election Simulation
Reproducible multi-agent simulation for studying opinion dynamics and LLM-agent behavior through controlled research runs.
- Python
- LLM Agents
- Simulation
- Research
Data & ML
Forecasting Oil Price Using AI
Research pipeline combining market data, news sentiment, and time-series models to evaluate forecasting experiments.
- Python
- Time Series
- NLP
- Machine Learning
Data & ML
Persian Text Summarizer
Flask API for Persian abstractive summarization with configurable generation, evaluation utilities, and explicit serving limitations.
- Flask
- Transformers
- Hugging Face
- Persian NLP
Experience context
Product work across software layers
Engineering approach
Designing for operation, not only implementation
- Reliability
Failure modes and recovery paths are part of the design.
- Observability
Systems should explain their state through useful logs and signals.
- Recoverability
Durable state is preferred over fragile memory-only workflows.
- Security boundaries
Trust, credentials, permissions, and external actions stay explicit.
- Testability
Components and integrations should be verifiable in isolation and together.
- Operational documentation
Deployment and troubleshooting deserve the same clarity as development.
- Honest limitations
Implemented behavior is separated from aspirations and unsupported claims.
Contact
Explore the work and current product context
GitHub contains the public repositories and technical documentation. Takaaneh provides current product context.