Autonomous AI · Deployed SaaS · Developer Tools — built from scratch, running in production.
Autonomous AI systems and automation platforms — purpose-built to eliminate cognitive and creative bottlenecks. Every system here is deployed, solves a specific problem, and handles real workloads.
If it doesn't run in production, it's just an essay.
| 🎮 Game shipped to Steam | Cropocalypse — 3D cell-shaded Unity/C# farming sim, publicly listed and available on the Steam Store |
| 🌐 SaaS serving real users | VocalCleaner — dual-engine AI audio mastering, live on Render and accessible to anyone right now |
| 🔧 Zero-install developer tool | PyAuditor — 25 static analysis rules, pure Python stdlib, scans thousands of lines in under 2 seconds |
| 🧠 39-module autonomous agent | JARVIS — self-healing cognitive OS with episodic memory, live skill synthesis, and zero-intervention failure recovery |
| 📊 Multi-modal virality platform | YouTube Analyse Pro — CV + audio AI + multi-agent debate engine, delivered through an encrypted multi-user SaaS |
The two most architecturally significant projects in this portfolio. Everything else builds around them.
Problem: AI assistants are demo-grade chatbot wrappers — they hallucinate, forget context, and break silently with no recovery path.
Solution: A 39-module cognitive agent OS with episodic vector memory (ChromaDB), a reflection engine that rewrites its own plan on failure, and a live skill synthesizer that generates and hot-loads new Python tools using AST sandboxing.
Impact: A fully self-healing autonomous agent — voice, GUI, and CLI inputs all routed through the same async orchestration core — that operates and recovers from failures with zero human intervention.
Real-world signal: Autonomously detects mid-task failures, rewrites its execution plan, and completes the goal — without a single human prompt, ever.
Architecture Highlights
- Episodic Vector Memory — recalls past successes and failures via ChromaDB embeddings
- Reflection Engine — detects plan failure, reasons about the cause, and re-plans autonomously
- Dynamic Skill Synthesizer — generates new Python tools on the fly, validated through AST sandbox before hot-loading
- Semantic Router — routes simple commands locally in milliseconds; escalates only complex cognition to the LLM
- Multi-input Core — Whisper STT (voice), GUI, and CLI feeding a unified async orchestration pipeline
Problem: Content creators optimize by instinct — no quantitative, reproducible framework exists for diagnosing why a video goes viral.
Solution: A dual-component platform (Desktop App + Chrome Extension) that uses OpenCV frame analysis, Librosa audio intelligence, and a multi-agent AI debate to produce a weighted viral DNA score for any video.
Impact: Turns any YouTube video into a reproducible viral blueprint — delivered through an AES-encrypted, Google OAuth2-authenticated SaaS with PDF report generation.
Real-world signal: Produces a full viral DNA report — frame-level cut analysis, audio tempo mapping, multi-agent hook debate, branded PDF — for any YouTube URL in under 60 seconds.
System Highlights
- Computer Vision (OpenCV) — frame-by-frame scene cut and pacing analysis
- Audio Intelligence (Librosa) — tempo, dead air, and decibel peak mapping
- Multi-Agent AI Debate — two AI personas argue to produce the strongest viral hook
- BabaClutch Chaos Metric — custom NLP algorithm measuring psychological engagement density
- DNA Scoring Engine — Hook 40% + Tempo 40% + CTA 10% + Emotion 10%
- AES-encrypted SaaS — Google OAuth2, multi-user, PDF export
Every project below is production-grade, purpose-built, and part of a coherent body of work.
Tools that save engineering time with no setup overhead and no excuses.
| Project | Description |
|---|---|
| PyAuditor | 25-rule AST static analyzer + BYOAI logic-review prompt generator. Zero dependencies. Pure Python stdlib. Sub-2-second scans. |
A complete creator pipeline — from virality analysis to rendered output, fully automated.
| Project | Description |
|---|---|
| RankVibe Automation | Autonomous short-form video generator — concept to rendered MP4 using AI + computer vision |
| ViralCut Pro | Zero-reencoding video trimmer — instant cuts via FFmpeg stream-copy and async byte streaming |
| AutoCutter | AI multi-track silence remover — Silero VAD, N-track simultaneous analysis, EDL output for DaVinci Resolve |
| VocalCleaner 🌐 | Dual-engine AI audio mastering — CleanVoice de-noise → Auphonic broadcast pipeline, live on Render |
Beyond content — production web infrastructure for real teams and real workflows.
| Project | Description |
|---|---|
| Finish & Click | Enterprise project management SaaS — Kanban + table view, Supabase RLS, real-time collaboration |
| Deal Hunter 🌐 | Real-time Steam & Epic Games discount aggregator — glassmorphism UI, price alerts, TR pricing |
| SuperArmor PDF Generator | Serverless AI exam prep PDF generator — Groq Vision OCR + Llama 3.3 70B |
One project built for the craft of shipping something players can hold.
| Project | Description |
|---|---|
| Cropocalypse 🎮 | 3D cell-shaded farming simulator — Unity/C#, automated agricultural pipelines. Listed on Steam. |
Most automation targets the obvious. I target the edges — cognitive workflows, creative pipelines, and systems complex enough that "no good tool exists yet" is the actual starting condition.
No scaffolding projects. No tutorial clones. Every repo here solves a problem that was too specific for off-the-shelf tools or too complex for a simple script — shipped with real architecture: memory systems, multi-agent pipelines, async orchestration.
Football sharpens the discipline. Software sharpens the mind.
Backend & AI Orchestration
Frontend & Design
AI, Vision & Audio
Infrastructure & Data
Open to high-signal conversations: autonomous AI research, architectural collaboration, or anything ambitious.
- 🐙 GitHub: @oguzemirtopuz
- 📧 Email: oguzemirtopuz10@gmail.com
- 🌐 Website: oguzemirtopuz.github.io