paritosh.space

Paritosh Varshney · Backend & Agentic Systems Engineer · Bangalore, India

I build the backend that keeps AI products running.

3+ years building high-throughput backend systems, and lately the orchestration and cost tracking behind production LLM features.

GitHubLinkedInTwitterEmail
Ask me anythinglive agent · soon
>
orchestrator — simulated run
$orchestrate meal-plan --days 7
✓fetch profile
✓generate day 1–7
!validate nutrition gates
✓retry day 4 · attempt 1
✓persist plan · notify app
▸done in 3m 28s · 128,455 tokens · $0.41

Selected work

What I've built recently.

django-admin-mcp

django-admin-mcp

MCP · access control

Read-only Django admin for AI agents

A drop-in Django package that exposes admin data to AI agents over MCP. Every read runs through the admin's own permission checks, secret fields come back redacted, and queries hit read replicas with timeouts. In production it cut debugging sessions from hours to minutes and lets agents triage customer support tickets.

Python · Django · MCP · OAuth PKCE

Code In production
faultline-orchestrator

faultline-orchestrator

Agentic infra · durability

Durable multi-step LLM agent workflows

An orchestration engine for long-running LLM agent workflows: validation gates, self-healing retries, and resume-after-failure so a multi-step run survives crashes instead of starting over.

Python · LLM Orchestration · Durability

Code Live demo in progress
multi-llm-web-orchestrator

multi-llm-web-orchestrator

Multi-provider · tooling

One prompt, four models, side by side

A Chrome side-panel extension that orchestrates ChatGPT, Claude, Gemini, and Grok through authenticated browser sessions — fan a single prompt across providers and compare answers in one place.

TypeScript · Chrome Extension · Multi-LLM

Code
mini-perplexity

mini-perplexity

RAG · retrieval

Search-powered, sourced Q&A

A compact Perplexity-style answer engine — runs live web search, grounds an LLM on the results, and returns concise answers with citations back to sources.

JavaScript · RAG · Search

Code

Experience

Where I've worked.

Healthify

NOW

Software Engineer 2, Backend

Dec 2024 — Present

  • Led backend for an insurance-backed telehealth platform from MVP to full US rollout — eligibility, provider discovery, booking, and cross-system insurance/account sync via Healthie, Stedi, and Stripe.
  • Built an LLM orchestration pipeline for multi-day diet-plan generation with parallel processing, validation gates, and self-healing retries — cutting turnaround from 25 min to 3.5 min.
  • Built centralized LLM token & cost observability across 50+ features — per-provider/model/feature breakdowns of prompt, completion, reasoning, cached, and audio tokens for budget monitoring.

Python · Healthie · Stedi · Stripe · MongoDB · MySQL · Redis

Metropolis Technologies

Software Engineer, Backend

Jul 2023 — Dec 2024

  • Architected an event-driven monitoring & alerting service for garage hardware, cutting incident response from hours to minutes.
  • Re-engineered the reservation subsystem to support 3,000+ garages and 2.5M+ annual bookings (100% YoY growth) while keeping the API flexible across configurations.
  • Refactored the payments microservice into a provider-agnostic architecture spanning Stripe, HPS, Square, and Chase.

Node.js · MySQL · MongoDB · MQTT · Microservices

Texas Instruments

Software Engineering Intern

Jul 2022 — Dec 2022

  • Built 35+ hardware-interfacing APIs across two generations of FMCW radar.
  • Optimized radar logic to cut non-sampling time by 30%, improving object-detection accuracy.
  • Delivered core integrations for global partners including Google, Ford, and Infineon.

C · C++ · Python

About

A bit about me.

I'm a backend engineer working mostly on the parts of AI that keep it running in production — orchestration, retries and durability, and tracking what the models cost. Most of what I do sits between systems engineering and LLM features.

Before the AI work I spent a couple of years on high-throughput systems: payments across four gateways, reservation infrastructure handling millions of bookings, and event-driven monitoring. That's where I learned how things behave under load.

I use coding agents to move quickly, and I rely on that systems experience to make sure what ships actually holds up.

Languages

Python · JavaScript · TypeScript · C · C++

Backend & Infra

Node.js · Django · Celery · Microservices · Event-Driven · AWS

Data & Messaging

MySQL · MongoDB · Redis · RabbitMQ

AI

LLM Orchestration · Agents · Token Observability · RAG · Prompt Engineering

Education

BITS Pilani, Goa

B.E. (Hons.) Electronics & Instrumentation, Minor in Physics

2019 — 2023

Community

Workshop & Operations Head, Quark 2022 (BITS Goa) — led a 180+ member team across workshops, partnerships, and outreach.