I design and ship AI systems end to end, from architecture to live product.
MSc Computer Science student at RPTU Kaiserslautern-Landau, Germany. Open to AI working-student roles (up to 20 h/week): LLM applications and agents, retrieval and ranking, computer vision, applied ML research.
Portfolio with architecture write-ups: https://goutham-dechineni.vercel.app
Tellability, live at tellability.com. An AI hiring platform I designed and built solo. Job seekers build their profile in a chat with an LLM agent; recruiters get candidates ranked by vector search, a structured score and a budgeted LLM review. In an early sample, Gemini context caching served 87.5% of prompt tokens from cache. The code is private; the write-up shows the architecture.
| Project | What it does | Write-up |
|---|---|---|
| codebase-mcp | MCP server that gives Claude Desktop local semantic code search, persistent memory and a quality-checked edit pipeline | architecture |
| ogphotos | Desktop app that cleans up a Google Takeout photo export on your own machine: duplicates, similar shots and faces | architecture |
| heart_risk_detection_project | Cardiovascular risk from a retinal image plus clinical data, fused into one result with an LLM explanation | architecture |
| Image_denoiser | Identifies the noise type, then routes the image to a denoiser trained for that type | architecture |
| TRANSFUZZY | Matches personal names across Latin and Indian scripts | architecture |
| fake_job_detector | Checks a job posting for scam signs and drafts an ATS-friendly resume for legitimate ones | architecture |
Other client systems (drug-interaction prediction with graph neural networks, CCTV event detection, resume screening) are private. Their write-ups are on the site.
- LLM apps and agents: Gemini, MCP, tool calling, context caching, Pydantic structured output
- Retrieval: pgvector, HNSW, OpenAI embeddings, MiniLM, EmbeddingGemma
- Computer vision: U-Net, ResNet, EfficientNet, YOLOv8, MediaPipe, InsightFace, EasyOCR
- Graph and classical ML: GATv2, RDKit, XGBoost, scikit-learn
- Backend and infra: Python, TypeScript, FastAPI, Next.js, PostgreSQL, Redis, Cloud Run, Vercel, GitHub Actions