Switchboard
Real-time voice-AI orchestration engine
Routes phone calls (PSTN/SIP) through STT → LLM → TTS with streaming and barge-in. Multi-tenant, sub-300ms turn latency.
- TypeScript
- Node.js
- WebSockets
- Redis
▪ p50 turn latency ≈ 280msOPEN →
SYS.01 — CONSOLE30.9°N · 75.8°E · STATUS NOMINAL
TUMPREET
SINGH
Backend-focused full-stack engineer · AI/ML
I build the real-time backend systems and the ML that make AI products actually work.
→ read more / about
↓ scroll — the console is real. Type help and hit ↵.
singh@console — /home/tumpreet
console v2.6.0 — deep-space link established
auth: Tumpreet Singh · Backend-focused full-stack engineer · AI/ML
┌────────────────────────────────┐
│ singh@console │
│ a deep-space developer console │
│ systems · realtime · ml │
└────────────────────────────────┘
I build the real-time backend systems and the ML that make AI products actually work in production — plus the tooling around them.
type 'help' to list commands — enable JavaScript for the live console.
singh@console:~$
Systems
3 FEATURED / 6 TOTALPrism
Transcript intelligence service
Real-time intent, sentiment & entity extraction over call transcripts. TensorFlow → ONNX with dynamic request batching.
- Python
- TensorFlow / Keras
- FastAPI
- ONNX Runtime
▪ batched inference < 50ms · intent F1 ≈ 0.9xOPEN →
Conduit
Integration middleware framework
A tenant / adapter / router framework to wire CRMs, telephony and webhooks together without per-client spaghetti. Dual deploy (SaaS + self-hosted).
- TypeScript
- pnpm monorepo
- PostgreSQL
- Redis
▪ idempotent webhook pipeline · replay + mergeOPEN →
Backend-focused full-stack engineer who also ships AI/ML (Python, TensorFlow). I build the systems and the tooling around them.