Senior Software Engineer & AI Architect in Switzerland. 14 years shipping software where failure is expensive — native C++/CUDA inference runtimes, enterprise AI agent platforms, and resilient off-grid packet-radio systems. Every number below ships with its script and raw data.
Sixi Lab (sixi.ai) · Portfolio · Writing (radeksixi.github.io) · Work with me · contact@sixi.ai · X · Open to remote roles and contracts.
1. sixi-asi — Hybrid MoE & Gated DeltaNet inference engine
C++20/CUDA speculative-decoding server for hybrid full-attention + Gated DeltaNet + sparse MoE models on NVIDIA Jetson Thor and DGX Spark unified memory, built on llama.cpp with zero Python on the runtime path (v1.0.0-rc1 — 1.0.0-rc1 write-up, 0.11 write-up, 0.9 write-up, 0.8 write-up).
| Stage | What it does |
|---|---|
| Verify & Plan | Verifiable .sixi manifest + memory planner that locks exact VRAM/RAM budgets up front (< 20 GiB for 27B hybrid, < 12 GiB low-VRAM, < 80 GiB for 512-expert MoE) |
| Accelerate | Lossless speculative decoding with O(1) recurrent-state rollback, block drafters, and a 106-line Thor kernel patch: 45.3 tok/s single-stream on Qwen3.8-27B (3.8× AR, level with TensorRT Edge-LLM NVFP4 on plain GGUF) |
Scale (v0.11) |
Input-log replay instead of 144 MiB state snapshots per row: 32 concurrent streams at 170 tok/s in 27 GB (down from 70 GB); 16 long-prompt agents wait < 3 s instead of 442 s |
Download & run (v1.0.0-rc1) |
One ready-to-run archive for the Jetson AGX Thor (download): ./sixi pull && ./sixi chat, an OpenAI-compatible server that installs as a background service, JSON answers constrained by a schema (120 of 120 valid against 103 without). Measured on the archive: 44 tok/s single-stream (llama.cpp v1.0.0 with built-in MTP drafts: 29), 171 tok/s across 32 streams |
| Remember & Serve | Cross-request prefix cache (1.0 s vs 209 s follow-up at 63k context), OpenAI-compatible HTTP/SSE server, native Qwen3.8 template, reasoning budgets, and schema-checked tool calls |
2. laya-as-judge — Calibrated AI evaluation & routing
Calibrated typed-decision evaluation engine (v0.4.0 Laya-Omega, Python · CUDA / MLX) that replaces uncalibrated single-LLM judges with fast local GPU encoders + selective cloud escalation (write-up).
| Layer | How it works |
|---|---|
| Local GPU Encoders | Laya + Vega-U 0.8B with disjoint Test-Time Training handle clear decisions locally in 7–14 ms at $0.00 |
| Selective Cascade | Escalates only uncertain or multimodal cases to packed Gemini Flash-Lite / Flash batches |
| Meta-Calibration | 10-fold out-of-fold simplex probability calibration produces honest confidence scores instead of raw model logits |
| Measured Result | 11/11 benchmark wins over Gemini Decision 1 across 2,250 questions, pooled Top-1 ECE 0.0234 (vs 0.0250), $0.77 total cost |
3. sixi-scanner — Agentic AI red-teaming & benchmark
On a real enterprise agent, 70–97% of generic red-teaming flags were false alarms. I build a static Go scanner and a wire-level benchmark that separate real violations from noise (v0.9.1 write-up, benchmark write-up).
| Project | What it does |
|---|---|
sixi-scanner · Go (v0.9.1) |
Static CLI scanner: 27 single- and multi-turn probes, zero dependencies, no LLM inside by default. 1st of 7 tools on precision (0.821) and recall (0.571), 1.000 precision (22/22) on multi-turn sessions; native SARIF 2.1.0 CI gate (exit 3 on partial coverage) |
| agent-redteam-benchmark · Python | Seven tools (garak, promptfoo, DeepTeam, PyRIT, Azure Red Team, agent-probe, sixi-scanner) vs a live Microsoft Foundry agent, scored from the wire by 10 deterministic oracles + a tool-blind judge |
| agent-redteam-labs · Python | Hands-on agent red-teaming and hardening labs + adk-demo-target, plus upstream A2ATarget merged into Microsoft PyRIT |
Rifugio & Swiss MeshCore — Off-grid field systems at sixi.ai
At Sixi Lab (sixi.ai) I engineer Rifugio — autonomous solar emergency infrastructure for Swiss MeshCore LoRa networks — alongside an open-source radio stack and live Swiss topology map:
| Project | What it does |
|---|---|
| Rifugio · Sixi Lab | Autonomous off-grid MeshCore emergency alert, help-request, and link-health programme for mountain valleys, with open specifications and radio bench reports |
sigillo · C99 · Python · TS (v1.1-rc) |
Fits an Ed25519-signed EU CAMF public emergency alert into 87 bytes for LoRa radio; verifiable 100% offline, with shared test vectors and receiver profiles (write-up) |
sixi-tdeck-plus (Faro) · C/C++ · LVGL 9 (v0.12.1) |
Open-source touchscreen/keypad UI and Ubuntu console across 5 ESP32-S3 LoRa boards (T-Deck Plus/MAX, ThinkNode M9, T-Beam Supreme, Heltec V3), offline maps, EN/IT/DE/FR/PL, and WebSerial flasher (write-up) |
| sixi-observer & GlassBoxEdge | USB LoRa / RTL-SDR receiver bridge feeding map.sixi.ai, upstream MeshCore PRs, and air-gapped Jetson edge diagnostics |
- Focus: Local C++/CUDA LLM runtimes (sixi-asi) · Calibrated hybrid AI evaluation (laya-as-judge) · Agent red-teaming & identity/governance on Microsoft Foundry & Google ADK (sixi-scanner) · Off-grid LoRa mesh & embedded firmware (Sixi Lab).
- Stack: C/C++ · CUDA · Python · Go · TypeScript · C#/.NET · ESP32-S3 / LVGL / LoRa (MeshCore) · llama.cpp · NVIDIA Jetson Thor & DGX Spark · Azure / GCP / AWS · MCP · A2A.
- Certified: Microsoft Multi-Agent AI Solutions Expert (AI-500) · Azure Solutions Architect Expert · Azure Security Engineer · AWS Security Specialty.