Live productGledva
AI Travel Companion · B2C
An intelligent travel workspace: AI itinerary planning, trip timeline, bookings, documents, tasks, packing, budget, reminders, weather, maps, nearby discovery and in-trip assistance across iOS, Android and Web.
What I owned: Product strategy · UX · requirements · AI architecture · prioritization · launch · GTM · analytics
- 100+ active users in Q1
- 3 platforms
- 0→1 launch
Live productTripHub
AI SaaS for Travel Advisors · B2B
Advisor workspace covering inquiry → traveler → proposal → trip → booking → documents → tasks → communication → completion — with a cited Knowledge Hub over supplier contracts, a WhatsApp and e-mail inbox with AI-drafted replies and client memory, and AI governance each agency controls.
What I owned: Product strategy · roadmap · UX · AI workflows · SaaS pricing · onboarding · GTM
- 20+ registered advisors
- 100% answer accuracy in eval
- WhatsApp + e-mail inbox
Live productGledMD
AI Medical Education
USMLE QBank, AI mock interviews, personal statement builder, ERAS CV review, school finder, interview readiness score and IMG Hub in one platform.
What I owned: Product strategy · requirements · AI experiences · packaging · pricing · iteration · GTM
- 50+ active users in Q1
- 0→1 launch
- Continuous validation
Architecture & prototypingVoice AI
Real-Time Voice AI · Systems Architecture
A real-time voice assistant for sales conversations, rebuilt from measurement: pipelined streaming instead of six sequential steps, memory-bandwidth-driven GPU choice, container sizing for cold start and cost, and sentence-level streaming to hide synthesis latency.
What I owned: Systems architecture · latency profiling · unit economics · hardware selection · streaming design
- 8.5s → 0.8s by GPU choice
- 87% of cost is voice
- 6 sequential steps pipelined
BenchmarkGPU Benchmark
Voice Inference Benchmarking · MI300X vs GH200 vs H100 vs Managed APIs
A full speech-to-text, LLM and text-to-speech pipeline on one self-hosted AMD MI300X, NVIDIA GH200 and NVIDIA H100 PCIe, measured against Azure AI Foundry and OpenAI on time to first audio, concurrency against a 500 ms SLO, shared-GPU isolation and cost per token.
What I owned: Benchmark design · latency SLOs · capacity planning · unit economics · ROCm/CUDA vLLM serving
- 331 ms to first audio (GH200)
- 256 calls per GPU (MI300X)
- ~$0.08 vs $2.08 per 1M tokens
Past experienceIntel
Connected PC & Wireless
5G/LTE connected PCs, WWAN, Windows drivers, PCIe, MBIM, IMS/VoLTE and hardware/software integration across the OEM ecosystem.
What I owned: Product management · technical leadership · OEM programs · performance and reliability KPIs
- 100+ design wins
- ~6 Gbps performance
- $10M+ recurring revenue