I am Pablo M. Suarez, founder of Apohara. I build silicon-native infrastructure for multi-agent LLM systems, validated on real AMD Instinct MI300X hardware. Every numeric claim I publish traces back to a peer-reviewed result or a reproducible benchmark — no exceptions.
🏛 Company Apohara, founder
🔭 Building Silicon-native KV-cache infrastructure for multi-agent LLMs
⭐ Flagship ContextForge — Zenodo DOI 10.5281/zenodo.20277875
💬 Ask me about vLLM, ROCm, AMD MI300X, multi-agent orchestration, MCP
🧪 Method Scientific. Logic over speed. Confirm or deny.
⚡ Fun fact Every number I ship has a DOI or a repro script behind it.
GitHub embed snippet
[](https://tokscale.ai/u/SuarezPM)I orchestrate fleets of coding agents — this is what the bill looks like.
| Project | What it does | License |
|---|---|---|
| ⭐ ContextForge | The first formally-verified safety layer for multi-agent LLM pipelines. Share the KV-cache to make agents affordable — without silently corrupting your judges. 44% fewer prompt tokens, 3.55× KV-cache VRAM reduction at INT4, 0/1210 safety violations. Carries a permanent Zenodo DOI. Flagship. | Apache-2.0 |
| 🚀 apohara-synthex | Evidence layer inside Bright Data. Scrape, classify, then sign verifiable web intelligence. MCP companion to brightdata-mcp. | see repo |
| Apohara-Catalyst | Rust local-first orchestrator that races the AI coding CLIs you already pay for. Parallel dispatch in isolated git worktrees, with quality gates. | see repo |
| apohara-consilium | Agent Governance OS for autonomous AI in regulated industries. | Apache-2.0 |
| apohara-probant | A different AI reviews the code your AI just wrote, while agent memory stays isolated. Defense-in-depth for AI-generated code. | see repo |
| apohara-aegis | Defense-in-depth trust layer for multi-agent LLMs. Governed context in, inspected prompts out. | Apache-2.0 |
| Apohara-Guard | Isolation primitives and ML-subprocess sandbox patterns for the PROBANT ecosystem. | AGPL-3.0 |
| apohara-framework | Private / in development. | — |
| rag-from-scratch | RAG implemented from scratch in Python, with step-by-step explanations. | see repo |
| brightdata-mcp | Fork of brightdata/brightdata-mcp. Not my project — PR opened upstream. | upstream |
The pitch is the curve, not a single number.
🪨 Silicon-native, or it doesn't ship.