Full-stack & AI developer · Mumbai, India

Parth Sangani

I build multi-agent AI systems and real-time products.

Full-stack applications, autonomous agent pipelines, and machine-learning systems, designed around clear user needs and built to hold up in production.

Internship: Business & Technology Analyst, Adbureau (May – Aug 2026)
Studying: B.Tech Computer Engineering, MPSTME
Open to: Software engineering & technology analyst roles

View selected work Download résumé Get in touch

Experience

Building an autonomous B2B intelligence platform end to end

Business & Technology Analyst Intern

Adbureau Analytics Pvt. Ltd., Mumbai, India

May 2026 – Aug 2026

PythonLangChainGoogle GeminiPlaywrightFastAPIPostgreSQLReact

Architected a multi-agent Company Intelligence Platform in Python that autonomously discovers, profiles, and scores B2B prospects as a 3-agent sequential hierarchy using LangChain and Google Gemini.
Developed a discovery agent integrating the Google Custom Search API with platform fingerprinting to return 20+ validated domains per keyword, plus a Playwright crawler that uses LLM prompting to convert unstructured HTML from 4 page types into structured JSON profiles.
Implemented a 100-point weighted scoring rubric and automated outreach generation, persisting records to a PostgreSQL schema served through a FastAPI backend with a React dashboard and CSV export.

Freelance

Client work shipped end to end for a cloud kitchen and a content agency

Live sites built and delivered for real clients, from payments and order handling to media-heavy front ends.

Fusion Feast

Mumbai cloud kitchen

Oct 2026 – Present

An end-to-end food-ordering platform with recurring meal-plan subscriptions, built so the owner can run it from a Google Sheet.

Next.jsTypeScriptRazorpaySupabaseGoogle Sheets APIWhatsApp Business API

Live site

Shipped the ordering platform in Next.js and TypeScript, supporting one-off orders and recurring meal-plan subscriptions with custom delivery scheduling.
Integrated Razorpay payments with signature verification and idempotent webhook confirmation, alongside Cash-on-Delivery, with prices recomputed server-side to prevent client-side tampering.
Used Google Sheets as the owner's live order database and Supabase (Postgres, Auth, Row Level Security) for customer accounts, kept in sync by server-side writes.
Automated WhatsApp order notifications for confirmations and status updates.

Vitra Media

Content agency

Sep 2026

A fast, theme-aware portfolio site for a content agency, with a content pipeline the team can extend without touching code.

ReactViteTailwind CSSFramer MotionLenisJSON SchemaVercel

Live site

Designed and built the client-facing site in React and Tailwind CSS, deployed on Vercel.
Optimized video delivery (faststart remux, poster frames, viewport-gated playback) for fast first paint on media-heavy pages.
Kept all case-study copy and media in JSON files validated against JSON Schema at build time, so a malformed entry fails the build instead of shipping a broken page.
Built light and dark modes by re-pointing a single set of CSS design tokens, with a saved choice, an OS-preference fallback and no flash on first paint.

Selected work

Projects across computer vision, real-time systems, applied AI, and mobile

A focused set drawn from coursework, hackathons, internships, and independent builds, each shipped to a working state.

MARD: giving recursive LLM calls a map of the document

Aug 2026 – Oct 2026

A metadata envelope that makes long-document reasoning with Recursive Language Models cheaper, more stable and traceable.

PythonOpenAI API (GPT-5.2 / GPT-5-mini)PydanticRecursive Language ModelsLaTeX

Case study Code on GitHub

Recursive Language Models let an LLM read documents far larger than its context window by exploring them through code, but every sub-call receives a raw slice and nothing else.
Co-authored Metadata-Augmented Recursive Decomposition (MARD), which passes a small envelope (document skeleton, accumulated findings, parent directive) into every sub-call. A frontier Scout produces a schema-validated concept graph, and a parallel budget-model Swarm builds dependency-ordered study material from it.
On a 939-page textbook MARD used 15.8× fewer input tokens (20.6× on a second book) and produced a repeatable structure of 83–84 concepts, where the vanilla RLM baseline varied from 75 to 190 across three identical runs.

ChatScale: Why Real-Time Apps Break When You Scale Them

Sep 2026 – Oct 2026

A Discord-style chat app, the continuation of my earlier real-time chat project, built to study what happens to WebSocket applications when you run more than one copy: reproduce the failure, fix it with Redis, and measure which signal should trigger autoscaling.

Node.jsReactSocket.IORedisPostgreSQLDockerKubernetesOpenShiftPrometheusGrafanak6

Case study Code on GitHub

Run a Socket.IO chat app as multiple backend copies and messages silently disappear, because each server only knows about its own connections. CPU, the default autoscaling signal, also says little about WebSocket load.
Reproduced the message-loss bug, then fixed it with a Redis pub/sub adapter in a 57-line change switched on by one environment variable. Containerised the app (multi-stage, non-root images, health probes, graceful shutdown), deployed it to a 3-node local Kubernetes cluster (k3d) with Kustomize, and instrumented it with Prometheus and Grafana. Also deployed the full stack to the Red Hat OpenShift Developer Sandbox from a single 17-object YAML file.
Measured on the local k3d cluster: a CPU-based autoscaler never scaled (36% against a 60% target), while one on a custom connection-count metric added a pod in about 50 seconds. In k6 load tests, p95 message latency fell from 110 ms to 34 ms going from 1 to 4 pods, with no message loss.

Automatic Vehicle License Plate Recognition

Jul 2025 – Oct 2025

A four-stage computer vision pipeline for license plate detection and OCR under real-world conditions.

PythonOpenCVYOLOv8nEasyOCRSupabase

Case study

Reliable plate extraction in noisy real-world imagery requires more than detection alone, especially under varying lighting.
Engineered a four-stage computer vision pipeline using noise filtering, edge profiling, and adaptive thresholding to isolate number-plate regions, then combined YOLOv8n detection and bounding-box analysis with EasyOCR recognition.
Logged more than 1,000 plate entries to a Supabase (PostgreSQL) schema with confidence scores, creating a strong base for high-volume testing and validation.

AI-Based Learning Platform (Smart India Hackathon 2024)

Oct 2024

A real-time conversational learning tool for interacting with historical personalities.

FastAPIWebSocketNLPLLM Integration

Case study

Built for Smart India Hackathon 2024, where the team placed in the Top 5 among 700+ competing teams.
Lets a learner hold a real-time conversation with a historical personality, served through FastAPI and WebSocket services.
Generates questions adaptively as the conversation unfolds, using NLP and LLM integration, to check whether the learner has understood.
Adaptive question generation reached over 85% accuracy in evaluating user comprehension.
Tackles a real gap: students learn history better through dialogue than static text, but a tutor that responds quickly and also measures comprehension is hard to build.

Museum Assistant Chatbot

A real-time AI assistant for ticketing, navigation, and visitor support.

FastAPIWebSocketRAGSupabaseEmailJS

Case study

Built for museum staff who needed faster visitor support across navigation, ticketing and contextual questions, without scaling manual communication.
A real-time assistant on FastAPI and WebSockets, so visitors get answers as they ask.
RAG workflows ground answers in retrieved content, keeping guidance context-aware.
Integrated with Supabase for data, and automated email communication through EmailJS.
Reduced manual communication effort by 40% while making visitor guidance faster.

About

I build full-stack and AI products that stay fast, practical, and reliable

I'm a final-year Computer Engineering student at Mukesh Patel School of Technology Management and Engineering in Mumbai, and completed a Business & Technology Analyst internship at Adbureau Analytics. My work spans full-stack development, multi-agent AI systems, and machine-learning pipelines in Python, Java, JavaScript, SQL, React, and Node.js.

I'm strongest where product thinking meets engineering depth: REST API design, predictive modeling, and data analysis. I'm looking for software engineering and technology analyst roles where both matter.

Solve the actual problem, then make it quiet, quick, and dependable.

Education: B.Tech Computer Engineering, MPSTME, Mumbai (2023 – 2027)
Standing: CGPA 3.43 / 4.00 · expected graduation July 2027
Coursework: DSA, Algorithms, DBMS, Operating Systems, Computer Networks, AI, Cyber Security, Distributed & Cloud Computing

Technical skills