Software Engineer · AI Systems · Cloud-Native · Published Researcher
B.Tech CSE graduate (9.2 CGPA, Dean's List) specializing in agentic AI systems, LLM orchestration, and cloud-native engineering. Built production-grade systems across two onsite internships at Amdocs and vCommission. Published researcher in Computer Vision — Springer LNNS Vol. 1710, ICIVC 2025.
- Redesigned an AI case triage workflow into a 6-phase agentic architecture with 7 MCP servers and 12 rule files — 30% accuracy improvement
- Built a Kubernetes deployment agent (~65 MCP tools) automating 5G/4G Openet product installs across 9 products
- Built Aegis — a distributed LLM evaluation framework with multi-judge consensus, semantic scoring, and cost tracking to $10⁻⁶ USD resolution
- Published: Hybrid CNN-RNN Deepfake Detection Framework — doi:10.1007/978-3-032-10664-3
Dual-pathway LLM eval platform: local zero-dependency validator + async Celery-distributed telemetry client. Runs deterministic assertions, semantic similarity (all-MiniLM-L6-v2), safety scoring (unbiased-toxic-roberta), RAG grounding, and parallel multi-judge consensus pipelines with mean-score fallback. Cost aggregation to $10⁻⁶ USD, Plotly regression heatmaps, Slack/Discord threshold alerts.
FastAPI PostgreSQL Redis Celery Streamlit Docker
Python MCP server (~65 composable tools) automating the full Kubernetes lifecycle for Amdocs Openet Platform (5G/4G) and Charging stacks. Five-tier values merge, preflight CRD validation, async observability dashboard. Resolved VoltDB OOM, PVC corruption, and Gloo CRD ordering in production. Turned multi-day expert-dependent installs into single-operator deployments.
Python MCP Kubernetes Helm Docker RHEL Elasticsearch Consul
Offline-first mobile app for spaced-repetition learning. SRS algorithm, three study modes (MCQ / Flashcard / Scenario), AI card generation via Groq (llama-3.3-70b-versatile), context-aware deduplication, streak tracking, JSON export.
React Native Expo expo-sqlite Zustand Groq API TypeScript
Multi-stage NLP pipeline: Universal Sentence Encoder embeddings → FAISS cosine-similarity search (0.5 threshold) → K-Means clustering → Gemini 1.5 Flash narrative synthesis. Outputs structured JSON reports categorizing narrative gaps by story type.
Python Flask React 18 TensorFlow Hub FAISS scikit-learn Gemini
Real-time air quality and meteorological tracking across 34 Indian cities. FastAPI + APScheduler ETL pipeline, two-tier caching (99% API call reduction, 800ms → 50ms load time), SlowAPI rate limiting, Docker Compose.
React 19 FastAPI PostgreSQL APScheduler Docker Compose Nginx
Languages
AI / ML
Backend & Data
DevOps & Cloud
Hybrid CNN-RNN Deepfake Detection Framework Intelligent Vision and Computing: Proceedings of ICIVC 2025 Lecture Notes in Networks and Systems, Vol. 1710 — Springer Nature, 2026 doi:10.1007/978-3-032-10664-3