Sanchit Gupta

AI infrastructure engineer·LLM inference on GPU·Air-gapped deployments

~/whoami

bash

$ whoami

Sanchit Gupta
AI Infrastructure Engineer @ Neuralix.ai
B.Tech CSE, IIIT Lucknow '25

$ cat what_i_do.txt

I keep large language models running on GPUs
that have no internet connection.

$ ./highlights.sh

• Core engineer on EKAM AI — an indigenous Defence
  AI-as-a-Service platform under the MoD iDEX ADITI 2.0
  initiative, launched at the Chanakya Defence Dialogue 2025

• Model serving on vLLM and NVIDIA Triton across GPU nodes
  — continuous batching, KV-cache sizing, request scheduling

• Instrumented telemetry, profiled hot paths, tuned async
  execution and caching → cut average API latency by 40%

• Delivery into secure air-gapped and VDI sites, where every
  package and model weight has to be carried in offline

~/infra/request-path

bash

$ cat architecture.txt

                          ┌────────────────────────────────┐
      client  ───────────▶│   FastAPI gateway  ·  React    │
                          └────────────────┬───────────────┘
                                           │
                ┌──────────────────────────┼──────────────────────────┐
                ▼                          ▼                          ▼
       ┌─────────────────┐        ┌─────────────────┐        ┌─────────────────┐
       │  vLLM           │        │  NVIDIA Triton  │        │  Qdrant         │
       │  LLMs · SLMs    │        │  OCR · VLM      │        │  vector search  │
       │  cont. batching │        │  reranking      │        │  RAG retrieval  │
       └────────┬────────┘        └────────┬────────┘        └─────────────────┘
                └────────────┬─────────────┘
                             ▼
                    ┌─────────────────┐        ┌──────────────────────────┐
                    │   NVIDIA GPU    │◀──────▶│ Prometheus · Grafana     │
                    └─────────────────┘        └──────────────────────────┘

                  ── no egress · no package mirror · no second chances ──

~/stack

bash

$ cat stack.txt

serving      vLLM · NVIDIA Triton · Ollama · CUDA
models       LLMs · SLMs · VLMs · OCR · embedding · reranking
retrieval    Qdrant · RAG pipelines
backend      Python · FastAPI · PostgreSQL · Redis · C++
platform     Docker · Kubernetes · Linux · NGINX · Git
observe      Prometheus · Grafana · Loki · Tempo · OpenTelemetry
frontend     React · JavaScript

$ ls ~/projects

AI-inference/        self-hosted GPU platform on Kubernetes
url-shortener/       distributed, sub-100ms under load
AgroSmart/           soil and yield prediction, Django + ML
CampusConnect/       MERN admissions portal

$ echo $CONTACT

sanchitguptaghj@gmail.com