From requirements to production ops, all built solo
Spanning enterprise AI knowledge and content production, short-form video production, trading decision systems, and an LLM gateway. Each lives on its own subdomain — decoupled, independently iterated, running under real production workloads with real traffic.
Enterprise RAG SaaS. PDF, Word, web pages and any unstructured docs — parsed, vectorized, semantically searched, and chatted with via multiple LLMs, fully integrated with a custom CMS. Multi-tenant isolation, action-level RBAC, natural-language Q&A with original-text citations.
Visit rag.rdwk.com ↗Auto-publishing for industry news channels: daily RSS topic fetch, AI-rewritten scripts, voice synthesis / voice cloning, subtitle recognition, keyword-matched footage — one chain. Cloud / local dual engine, one-click switch — local mode runs digital avatars and text-to-video on your own GPU with on-demand VRAM scheduling. Zero human in the loop.
Visit reel.rdwk.com ↗AI trading SaaS: a multimodal LLM reads K-line charts directly, recognizes patterns and resistance levels, infers entry / stop-loss / tiered take-profit, then dispatches EAs to MT5 for fully unattended execution. No code, no quant background, 24/7 monitoring with emotions out of the loop.
Visit quant.rdwk.com ↗Self-hosted multi-model AI gateway with a unified OpenAI-compatible entry point, model switching, smart routing, and centralized call management. It serves as the AI backbone across RDWK products.
Visit relay.rdwk.com ↗Each project gets a deeper look at its core architecture, key technical choices, and engineering trade-offs, with flow diagrams mapping every stage. Every choice was shaped by real production constraints. Visit the matching subdomain for the live version.
Once PDF, Word, Excel, or web docs are uploaded, the RAG engine parses, chunks, vectorizes, and indexes them automatically, then returns answers with source citations. It is deeply integrated with a custom CMS and supports multi-tenant isolation, action-level RBAC, and multi-model Q&A for internal knowledge search, policy lookup, and document retrieval.
Built for industry news channels and content pipelines: it fetches RSS topics daily, rewrites scripts automatically, then chains voice synthesis / voice cloning, subtitle recognition, keyword-based media matching, and local/cloud rendering. The local and cloud engines switch in one click; local mode can also run digital avatars and text-to-video with on-demand VRAM scheduling, end to end without human intervention.
A multimodal LLM reads K-line charts directly, identifies patterns, resistance levels, and trend structure, then reasons about entry, stop-loss, and tiered take-profit before dispatching trade instructions to MT5 EAs. Cloud-side inference handles the heavy lifting; the local side only loads the EA, so the system can run 24/7 with emotions removed from the loop.
A self-hosted multi-model AI gateway with a single OpenAI-compatible entry point for model access, switching, and call management. Rag, Quant, and Reel all run on top of it, making it the shared AI backbone of the RDWK stack.
RDWK is my personal suite of AI products, covering enterprise AI knowledge and content production, short-form video production, trading decision systems, and a multi-model AI gateway. I entered the industry in 2015 and started independently building these projects in 2022. On each one, I personally handle requirements, architecture, code, deployment, and ops end to end.
What I care about is turning AI into engineering that can run in real businesses, not demos or pitch decks, but systems that have to deal with real data, real concurrency, and real outages. That's why every product runs in production, keeps iterating, and serves real use cases.
AI doesn't lack papers — it lacks people who turn models into products. RDWK is my record of that work, and I'll keep publishing sanitized project materials and engineering notes, along with the technical details and lessons learned.
Technical chats, opportunities, project discussions — all welcome via email