I lead the Python backend at a computer-vision product company — designing the async, streaming, and data layers that production AI workloads run on. By night, I write novels and cut short films.
The way I think about a production backend: keep the request path thin, push everything heavy off it.
flowchart LR
C(["Clients<br/>web and RTSP cameras"]) --> N["Nginx"]
N --> API["API layer<br/>Django and FastAPI"]
API -->|enqueue jobs| Q{{"Broker<br/>RabbitMQ and Redis"}}
API <-->|cache and sessions| R[("Redis")]
API <-->|read and write| PG[("PostgreSQL")]
Q --> W["Async workers<br/>Celery and ARQ"]
W -->|video and AI jobs| ML[["CV and video analytics"]]
W --> PG
W -->|artifacts| S3[("S3 object storage")]
MTX["MediaMTX<br/>RTSP to WebRTC"] -->|live feeds| API
API -.->|real-time| C
classDef accent stroke:#00d668,stroke-width:1.5px;
class API,W,MTX,ML accent;
Live camera feeds and AI analytics surface in real time, while video-processing and ML jobs stay off the request path. This is the shape of the Video Management System and Ikshana work I own at Intozi.
|
backends |
REST routes |
ORM tables |
Ansible roles |
KAVACH'23 |
Lead backend at a computer-vision & video-analytics product company. I own the Python services across the products and coordinate delivery within a small team.
Note
Currently building the backend for a client-facing Video Management System — Django services with MediaMTX wired in for RTSP/WebRTC, so live camera feeds and AI analytics surface in the app in real time.
- Own the Python services across the company's products, including Ikshana, its core video-analytics product.
- Architected the data & async layers —
PostgreSQL·Redis·RabbitMQ·Celery— to run video-processing jobs and AI workloads off the request path for responsive, scalable services. - Built an internal MLOps pipeline (Django + React): dataset upload → auto-labeling → human verification → training/retraining, with a path to client-facing deployment.
Four backends I designed and shipped end-to-end — deeper write-ups live in each repo.
| Project | What it is | Built with |
|---|---|---|
| Papyrus | Self-hostable, privacy-first PDF studio — merge / split / compress / OCR with a zero-retention TTL | FastAPI · Celery · S3 · Compose + Helm |
| HeadTogether | Geo-bounded, ephemeral chat rooms discoverable only by people physically nearby | FastAPI · WebSockets · Redis pub/sub · Argon2id |
| ShieldBuntu | One-click Ubuntu CIS hardening — apply / dry-run / revert, streamed live to the UI | FastAPI · Ansible · SSE · PAM |
| MeshHawk | Local-first 802.11 mesh detector — .pcap in, topology graph + SVG report out |
FastAPI · scapy · NetworkX · ARQ |
| Languages | Python · Bash |
| Frameworks | Django · FastAPI |
| Data & async | PostgreSQL · Redis · RabbitMQ · Celery · ARQ |
| Streaming & real-time | MediaMTX (RTSP/WebRTC) · WebSockets · Server-Sent Events |
| Infra | Docker · Kubernetes/Helm · Nginx · AWS · Linux · Git |
Engineering isn't the only thing I ship. I'm a published author — a novelette and two novels — and I shoot and edit short films. Both are the same discipline as backend work: structure, revision, pacing, and deciding what to leave out.
- 🏆 KAVACH'23 — Winner of the inaugural nationwide cybersecurity hackathon organised by the Government of India.
- 🎓 B.Tech, Computer Science — Bhilai Institute of Technology, Durg · CPI 9.68 (2020 – 2024).
-000000?style=for-the-badge&logo=x&logoColor=white&labelColor=00d668){kind=icon}
Thin request path. Heavy lifting off to the side. Same goes for the README — thanks for scrolling.