I build
From custom LLM microservices to automated ERP platforms — built end-to-end for real operational scale.
Systems designed, implemented, and maintained in live production — serving real users, handling actual load, and architected for high reliability.
An AI-native learning platform that turns student behaviour and academic data into personalized explanations and parent intervention reports. Runs a self-hosted Gemma 4B LLM (4-bit quantized) via llama.cpp and a Kokoro TTS microservice — zero third-party API dependencies, completely owned infrastructure.
A full-scale institutional academic platform featuring 70+ secure REST APIs, automated attendance tracking completed in under 60 seconds, role-based workflows, and an integrated autonomous assistant. Actively serving 1,700+ registered students and 50+ faculty members.
An internal Database-as-a-Service that provisions isolated PostgreSQL + MinIO instances for microservices in under 60 seconds. Designed with automated streaming standby replication, hourly liveness probes, encrypted secrets, and zero-touch crash recovery.
Industry-tailored ERP and quoting engine designed specifically for Indian UPVC fabricators. Features real-time BOM costing, branded customer portals, and multi-tenant mobile applications with automated CI/CD client APK builds.
Full-cycle technical ownership: from architecture and protocol design to infrastructure management and production reliability.
First engineer. Architected and deployed the core Nova My Mentor platform (custom LLM inference pipeline, TTS microservice, student analytics), Acharya ERP (70+ REST endpoints, 1,700+ users), and Nidhi DBaaS. Maintain a production VPS fleet running 15+ containerized services.
Bootstrapped and built a vertical ERP for UPVC fabricators. Designed multi-tenant architecture, created Flutter mobile applications with dynamic theming, and built the full Next.js cloud portal and database schema.
Specialized in distributed AI systems, neural inference optimization, and systems engineering. Continuously applied academic concepts directly into production environments.
Infrastructure must heal and deploy itself. Watchtower triggers automated container updates in ~30s upon CI build completion, while Nidhi provisions databases without manual intervention.
Rather than depending on costly external AI APIs with unpredictable latencies, I run local models (Gemma 4B via llama.cpp) and microservices (Kokoro TTS) directly on hardware.
Writing code is only 30% of the job. I monitor container memory thresholds, configure reverse proxies, manage streaming backups, and ensure uptime across all platform layers.
Available for engineering leadership, systems architecture discussions, and high-impact AI/platform collaborations.