AI Platform Engineer & Tech Lead
I build production AI systems and the platforms they run on.
RAG · vLLM inference · evals · agents over MCP, in Python, on Go & Kubernetes platforms I build myself.
whoami: ai platform engineer & tech lead · 7+ yrs. kubectl get pods -n ai-platform: RAG · inference · agents in production. cat ~/focus: reliability & cost, not demos.
About
I build platforms and the AI that runs on them
Senior engineer and tech lead who turns messy, real-world requirements into reliable production systems.
I'm an AI platform engineer and technical lead who ships production AI end to end. Most recently I co-founded Neosenth and shipped ViHi, a video-first services platform live on the App Store and Google Play, building its AI layer from scratch: RAG on pgvector, Gemini media analysis, and a Python ML ranking service.
I work mainly in Python for the operations layer around LLMs: RAG pipelines (chunking, hybrid search, reranking), self-hosted inference of open-weight models (Llama, Qwen, Mistral) on vLLM plus hosted Gemini, automated evals (RAGAS, Langfuse) wired into CI, LLM tracing and observability, and tool-calling agents over MCP. I care about reliability and cost, not demos.
The platform underneath is mine too: event-driven microservices in Go on Kubernetes (GKE), multi-cloud Terraform across GCP, AWS, and Cloudflare, with pgvector and Qdrant for retrieval and a full observability stack. Earlier, an ed-tech startup I co-founded earned recognition from India's Finance Minister, and at Zeliot I built real-time systems that process high-volume vehicle telemetry at fleet scale.
I've shipped directly for customers as a founder, turning messy real-world requirements into working solutions, and I've led teams of 10+ engineers. Whether it's an AI-platform, forward-deployed, or platform engineering role, I'm interested in building technology people actually use. Let's connect.
At a glance
- Scale
- 15+ Go microservices in prod · led 10+ engineers
- Focus
- RAG · vLLM · evals · agents over MCP
- Recently
- Co-founded Neosenth, shipped ViHi
- Based in
- Bengaluru, India
- Recognition
- Work recognized by India's Finance Minister
- Open to
- AI-platform · Platform / Backend · FDE roles
Experience
Where I've worked
From co-founding in 2019 to senior engineering and technical-lead roles across startups and enterprise.
Neosenth
Jan 2026 to Jun 2026Technical Lead & Co-Founder
Shipped ViHi from zero to live on the App Store & Google Play: built the AI layer end to end (RAG on pgvector, Gemini plus self-hosted vLLM, tool-calling agents over MCP, automated evals) on 15+ event-driven Go microservices. Mentored 5 interns.
- RAG
- vLLM
- MCP
- Python
- Go
- Kubernetes
Lumen Technologies
2024 to 2026Senior Software Engineer (SDC-II)
Built an internal LLMOps assistant that analyzes production logs and suggests grounded fixes: RAG over company docs with evals and guardrails. Also delivered Colorless and Intent-Based Networking automation with Temporal and Itential.
- RAG
- LLMOps
- TypeScript
- Temporal
Zeliot
2023 to 2024Senior Software Engineer (Team Lead)
Led 10+ engineers building SML Saarthi (fleet analytics for SML ISUZU): AI and computer-vision vehicle tracking sustaining ~400ms latency, plus GenAI/RAG-powered analytics.
- Computer Vision
- GenAI/RAG
- Node.js
- Kafka
S2T
2021 to 2023Software Engineer
Owned on-prem Kubernetes and led the Docker Swarm to Kubernetes migration for AI investigation products (WEBINT/OSINT); cut service latency to sub-second.
- Kubernetes
- AI Products
- Microservices
Career Capsule
2019 to 2021Technical Lead & Co-Founder
Co-founded an ed-tech startup; built a WebRTC live-class platform handling 100+ concurrent students per session.
- WebRTC
- Node.js
Recognized by India's Finance Minister
Skills
What I work with
Depth in backend platforms and the AI systems that run on them, not a checklist of everything I've touched.
My core is the AI operations layer, built mainly in Python: RAG pipelines (chunking, hybrid search, reranking), self-hosted inference of open-weight models (Llama, Qwen, Mistral) on vLLM plus hosted Gemini, automated evals (RAGAS, Langfuse) wired into CI, LLM tracing, and tool-calling agents over MCP. I optimize for reliability and cost, not demos.
It runs on platforms I build myself: event-driven microservices in Go on Kubernetes, provisioned with multi-cloud Terraform across GCP, AWS, and Cloudflare, with gRPC/ConnectRPC, Kafka and NATS, PostgreSQL and Redis, all wired into a full observability stack.
I've led teams of 10+ engineers, mentored 5 interns, and shipped products end-to-end as a founder, turning ambiguous real-world requirements into systems that run in production.
AI / LLM Platform
10The platform and operations layer around LLMs, built for reliability and cost, not demos.
- RAG: Retrieval-Augmented Generation: chunking, hybrid search, reranking.
- vLLM: Self-hosted inference for open-weight models (Llama, Qwen, Mistral).
- MCP: Model Context Protocol: connects agents to tools and data.
- Evals (RAGAS · Langfuse): Automated LLM evaluation and tracing, run as a CI quality gate.
- Langfuse: LLM tracing and observability: prompts, latency, tokens, cost.
- pgvector: Vector similarity search inside PostgreSQL.
- Qdrant: Vector database for similarity search and retrieval.
- Gemini: Google's hosted multimodal models, used via API alongside self-hosted vLLM.
- LangGraph: Stateful, multi-step tool-calling agents.
- PyTorch: Deep-learning framework underpinning model work.
Languages
5Day-to-day for AI services, platform code, and tooling.
- Python: AI services, FastAPI data layer, and ML ranking pipelines.
- Go: Event-driven microservices and platform tooling.
- TypeScript: Type-safe services and front-ends.
- JavaScript: The language of the web.
- Node.js: Runtime for scalable server-side apps.
Backend & Data
9Event-driven services, APIs, durable workflows, and storage.
- gRPC: High-performance, Protobuf-based RPC.
- ConnectRPC: Simple, Protobuf-based RPC over HTTP.
- GraphQL: Query language and runtime for APIs.
- FastAPI: High-performance Python API framework.
- Kafka: Distributed event-streaming platform.
- NATS: Lightweight, high-performance eventing (JetStream).
- Temporal: Durable, fault-tolerant workflow orchestration.
- PostgreSQL: Relational database (with PostGIS and pgvector).
- Redis: In-memory store for caching and queues.
Cloud & Infra
12Multi-cloud, containerized, and observable by default.
- Kubernetes: Container orchestration (GKE in production).
- Docker: Packages and runs apps in containers.
- Terraform: Infrastructure-as-code across GCP, AWS, and Cloudflare.
- Helm: Package manager for Kubernetes deployments.
- GitOps (FluxCD): Declarative, Git-driven continuous delivery to Kubernetes with FluxCD.
- GCP: Google Cloud, including GKE.
- AWS: Amazon Web Services cloud platform.
- Cloudflare: Edge networking, DNS, and Workers.
- CI/CD: Continuous delivery with GitHub Actions.
- Observability: Grafana, Loki, and VictoriaMetrics for metrics, logs, and dashboards.
- Linux: Unix-like operating system foundation.
- Git: Distributed version control.
Mobile
2Cross-platform clients for shipped products.
- Flutter: Cross-platform mobile app framework.
- Dart: The language behind Flutter.
Projects
Things I've built
Production systems shipped end-to-end, from AI platforms to enterprise telematics.
Video-first services platform: post a short video of your problem, get AI-matched professionals, compare bids, and hire. The AI layer is built end to end: RAG on pgvector, Gemini media analysis and liveness, tool-calling agents over MCP, and self-hosted vLLM, running on 15+ event-driven Go microservices on Kubernetes.
Writing
Notes from the build
Practical write-ups on backend platforms, AI in production, and scaling real systems.
Contact
Let's build something
Open to AI-platform, Platform / Backend & Forward-Deployed roles · Remote or relocation. Email is fastest.
Connect with me
Let‘s get in touch.