🚀 Building Hermetica: https://hermetica-ai.com/
🚀 Selected among 20 Greek founders worldwide for the first OpenAI accelerator: https://openai.com/global-affairs/openai-for-greece/
- Agentic Document Intelligence: Multi-agent pipelines for long-context reasoning over financial documents — extraction, cross-validation, gap detection.
- Advanced RAG: Hybrid retrieval (dense + sparse), reranking, citation grounding, hallucination guardrails, and eval loops purpose-built for high-stakes queries.
- LLM Orchestration: Stateful agent workflows with LangGraph; structured outputs with instructor/Pydantic; multi-model routing (GPT-5.4 mini, GPT - 5.4 nano, Gemini 2.5 flash).
- Product Engineering: clean domain-driven architecture, async-first backend design.
| Layer | Tools |
|---|---|
| LLM / AI | GPT-5.4 mini, GPT-5.4 nano, Gemini 2.5 flash, OpenAI Embeddings |
| Orchestration | OpenRouter, OpenAI Assistants API, custom agent loops |
| Document Processing | Docling, Unstructured.io, PyMuPDF |
| Retrieval | pgvector, Qdrant, Cohere Rerank, BM25 hybrid search |
| Backend | Python, FastAPI, Pydantic v2, Celery, Redis |
| Frontend | Next.js 15, React, Tailwind CSS, Vercel AI SDK |
| Infra & DevOps | Docker, AWS (ECS + RDS), GitHub Actions, Terraform |
| Observability | LangSmith, Helicone, Grafana |
| Evals | RAGAS, custom domain-specific eval harnesses |