Dhanush M
Software Development Engineer · Building production AI — voice agents, RAG, LangGraph workflows · Node · Python · AWS · Kubernetes
About
Software engineer building AI systems that run in production — with the authentication, queues, tenancy, monitoring and infrastructure underneath them. My work spans LLM enrichment pipelines for security operations, autonomous voice agents over real telephony, LangGraph agentic workflows on knowledge graphs, and retrieval-augmented systems in regulated domains. I build the platforms and I still operate them: Node and Python on the backend, React on the frontend, AWS with Terraform and Kubernetes underneath.
Skills
GitHub Contributions
Work Experience
Rootent TechnologiesFull-timeOn-site
Software Development Engineer
- Own the LLM enrichment pipeline for a security operations platform that classifies incoming alerts as true or false positives, covering the work queue, batched inference, retrieval-augmented context, threat-intelligence enrichment and delivery.
- Cut per-alert delivery latency by streaming verdicts out of each batch as they resolve instead of blocking on batch completion, and bounded generation cost with a caller-supplied output token cap.
- Built the graded benchmark harness the pipeline ships against, with accuracy, recall and latency gates enforced per run so prompt and model changes are regression-tested rather than eyeballed.
- Made the platform deployable three ways from a single Terraform module — managed SaaS, customer self-hosted and fully air-gapped — on Kubernetes with Elasticsearch semantic search and hot-swappable Gemini or local Ollama inference for data-residency-constrained customers.
- Built an autonomous voice AI screening agent that calls candidates, runs a structured interview and returns scored transcripts: four services (Node/Express, Python/FastAPI, React/Vite) streaming Plivo telephony → Deepgram STT → Gemini → Cartesia TTS over WebSockets with Pipecat, RabbitMQ between services, BullMQ/Redis queues and a full candidate state machine.
- Migrated a monolithic LLM workflow engine to LangGraph StateGraphs with router-based turn dispatch and state persisted across seven multi-turn flows, backed by a Neo4j cross-domain knowledge graph with embedding retrieval on Azure OpenAI.
- Built a RAG pipeline over SEBI ICDR regulatory templates using reasoning-based tree retrieval, served from a containerized FastAPI microservice behind a Node API.
- Deploy and operate the multi-tenant platform on AWS: subdomain → tenant → database routing, per-tenant nginx vhosts and TLS, pm2 supervision, and GitHub Actions CI/CD with rollback across three environments.
Roll2Bowl TechnologiesFull-timeOn-site
Founders Office · SDE Intern
- Built and shipped a customer-facing food ordering application using React Native, Node.js, Express and MongoDB.
- Developed a restaurant partner application with real-time order queue management over WebSockets and a reliable push notification pipeline.
- Contributed to the full-stack admin dashboard for order monitoring and analytics.
Projects
Meeting Notetaker Bot
Queue-driven bot that joins Microsoft Teams and Google Meet calls in a headless browser, records the session, and returns a transcript and summary. Handles mid-call drops with automatic rejoin and validates captured audio before inference to prevent hallucinated transcripts.
Docs
Collaborative text editor where multiple users edit documents with real-time updates and see active collaborators.
QuickPay
P2P money transfer app with Prisma for migrations and a Node.js/Express webhook handler for bank operations.
Medium
Serverless blogging platform on Cloudflare Workers with authentication, content management and edge-cached delivery.
DocPortal
Patient records platform with voice-to-text intake and LLM-generated formatted clinical reports. (code is not open sourced)