// Anirudh Purohit · Fullstack Developer · 2026

Anirudh
Purohit.

Fullstack developer experienced in product ideation, architecture, implementation, deployment, networking, and maintenance. Capable of building infrastructure from scratch and working within existing cloud environments.

View work ↓ anirudh.purohit@utdallas.edu
01 · Selected Work

Things I've shipped.

/01 Tasker

An NLP-powered task manager. Type a task in plain English; it parses intent, priority, color, and timing. 30+ users worldwide, 8000+ tasks processed.

↑ live widget · try the theme switcher

Open ↗
800/s
sustained req/sec
1.1s
LLM latency
8000+
tasks processed
Infra & scaling
  • ·Stateless Flask, horizontal scaling via containerized replicas
  • ·~800 req/sec sustained across a 2-node cluster via Caddy load balancer, weighted round robin 3:1
LLM integration
  • ·Latency-aware LLM gateway with live vendor health probing and adaptive failover
  • ·Cold-start latency cut from 10s+ to 1.1s p95 / 1.5s p99 via TCP pooling, custom keepalive, and prompt engineering
  • ·Parses NL into structured tasks; aggregates anonymized data for fine-tuning
Surface
  • ·Desktop and mobile PWA, 6 Jinja2 templates across 3 themes
  • ·Multi-platform push notifications, timezone-aware scheduling
  • ·Vanilla HTML/CSS/JS for maximum portability, no frontend framework
Flask PostgreSQL Docker Cloudflare Python Linux

Infrastructure

Two nodes: a late-2012 Mac Mini and a 2017 laptop with the battery yanked, both running Linux. Behind Cloudflare Tunnel, they host Tasker alongside everything else I'm building. It's my first stop for hosting anything new.

  • ·Custom CI/CD and provisioning scripts
  • ·Container-orchestrated via Docker
  • ·Horizontally scaled services across nodes
  • ·Zero-downtime rolling deploys
  • ·Health checks with auto-restart
  • ·Cloudflare edge: caching, rate limiting, WAF
  • ·Centralized logging
  • ·Multi-channel alerting
  • ·Automated DB backups with tested restore
  • ·Per-service status pages
/02

SmartAttendanceTracker

Attendance in large lecture classes is slow, fakeable, and a pain to consolidate. SmartAttendanceTracker fixes that with cryptographically signed QR tokens that rotate on a tight time window. Fast to take, hard to spoof, easy to consolidate.

Flask PostgreSQL HMAC-SHA256 Docker HTML/CSS/Vanilla JS
SmartAttendanceTracker live attendance dashboard

Enrollment happens once: students scan a QR the professor displays in class to join the course roster. From then on, the flow flips. Each student's device receives a fresh HMAC-signed QR every 15 seconds, and they walk past the professor's scanner to be marked present.

Plenty more on the roadmap if institutions greenlight (network checks, geofencing, device MAC-binding).

My role: ideated the project and led a 6 person team across 6 development phases.

/03

Sentience

HackAI UTD 2026 · 24hr

Market intelligence platform forecasting public sentiment toward AI labs and tracing supply-chain dependencies to predict downstream stock movement.

PyTorch FinBERT LSTM MongoDB Node Express React
Sentience dashboard

Custom parallel Python scraper ingests thousands of Reddit posts into MongoDB. ML pipeline: FinBERT scores sentiment (70% post + 30% VADER comment weighting), extracts 7 behavioral signals per post (churn, advocacy, trust erosion, competitor mentions, emotion intensity, engagement weight, comment alignment), aggregates into daily buckets, and trains a 2-layer LSTM to forecast composite sentiment 14 days out.

The differentiator: an interdependency network mapping AI companies to hardware suppliers like NVIDIA, TSMC, and AMD, so sentiment shifts at an AI lab get traced downstream to flag potential market movement before it happens.

My role: parallel scraper pipeline plus Express/Node REST API serving sentiment scores, forecasts, and supply-chain data to the React frontend.

SMC Inventory Management System
/04

SMC IMS

Client project · private

Full-stack inventory management for a medical equipment company. 185+ items, $1.3M+ inventory value.

Built primarily as an exploration of ETL pipeline design: extraction, transformation, and loading for inventory records including images, pricing, and manufacturer metadata. Role-based account permission layers, audit logging, and both direct and semantic search across item names, models, manufacturers, PI/CI references, and remarks.

Flask PostgreSQL Docker HTML/CSS/Vanilla JS ETL pipeline
02 · About

Let's build something.

I got into software because I wanted to understand how things actually work, not just use them. That's pushed me toward building end to end and keeping things running myself. I use modern tools freely, AI included, but try not to let them replace the thinking. Everything on this page is something I made and currently maintain.

GitHub
ani-pur
LinkedIn
ani-pur