building production-grade GenAI & systems
GenAI & Backend Systems Engineer
Low-latency RAG pipelines, LLM orchestration, Spring Boot microservices, and zero-downtime Kubernetes deployments — on infrastructure that runs at carrier scale at Biocipher Technology. Reliable AI in production, not API demos.
1.7B+
records optimized
100M/day
dial capacity designed
3+
AI products in production
5
engineers led
featured system
Natural-language-to-SQL over a 1.7B+ record enterprise store. An LLM planning layer receives the relevant schema context, generates validated SQL, and runs it against an index-tuned big-data tier — business teams pull answers without writing a line of SQL.
deployment pipeline
GitHub Actions workflows building and rolling microservices onto Kubernetes with health-gated rolling updates — voice traffic never drops mid-deploy.
zero-downtime updates
voice ops at scale
100M/day
dial capacity design
1.7B+
CDRs tracked
IVR·OBD·OTP
products live
weekly
release cadence
Asterisk & FreeSWITCH campaign infrastructure behind pixabits.ai — agentic voice calling, IVR trees, OBD campaigns, and voice OTP in production.
more builds ↓
A fully on-device Android AI assistant built with Kotlin and Jetpack Compose. Local LLM inference with zero cloud dependency, encrypted storage via SQLCipher and Android Keystore, and voice input. No analytics, no telemetry — privacy by architecture.
case study →
Optimized retrieval-augmented generation architecture: dynamic content-aware chunking, hybrid dense + sparse search, cross-encoder re-ranking, and low-latency vector retrieval for grounded, context-aware LLM responses.
case study →
Ground-up redesign of a CPaaS customer console in Next.js 15 and React 19. Visual IVR flow builder with nested DTMF trees, campaign composer, TTS template management, and agent-group routing — complex telecom workflows distilled into a clean, scannable interface.
case study →
Daily job-search platform for the Indian market aggregating Adzuna and Google for Jobs. Hourly-refreshed board, SMTP email digests, hand-rolled auth (email OTP, scrypt, HMAC sessions), and resume-based personalization with shared caching to stay inside free API quotas.
case study →
Enterprise-grade outbound-voice telecom platform architected for 100M dial attempts/day. Multi-node FreeSWITCH with Kamailio, channel-aware pacing in Go, PostgreSQL + ClickHouse CDR pipeline, and Kafka event streams — hexagonal, event-driven, horizontally scalable.
case study →
Windows desktop app in Tauri 2 + Rust that chats through a local Ollama LLM, manages tasks, and auto-writes a daily work journal from developer-tool transcripts on a nightly schedule. Fully local and free.
case study →
2026 — Now● current
Leading the voice marketing product and a team of 5 — AI agentic voice calling, IVR, and OBD campaigns shipped weekly on the CPaaS panel behind pixabits.ai. Cut customer resolution time 30% at a 95% on-time release rate.
2025 — 26
Tuned queries on a 1.7B+ record system, built Spring Boot APIs 15% faster with Redis caching, and kept production voice infrastructure up through critical incidents.
2024
Spring Boot REST APIs with JWT auth; Jenkins + Docker CI/CD pipeline that cut deployment time 10%.
2023
Responsive React components and client-side validation for an e-commerce platform.
3 — ● line open
A product to lead, a platform to scale, or a conversation about voice tech and on-device AI — my inbox is open.
vishalyadav68948@gmail.com