Pranav Ojha

Nova

What I’m building now

• TSTypeScript
• DBDatabases (Postgres)
• QATesting (Vitest)

Nova is a persistent personal delegate: tell it what you want taken care of, and its vision is to carry the task through, verify the outcome, and report back.

You describe an outcome in plain language, and Nova takes responsibility for seeing it through. It should plan the work, act only through capabilities you have authorised, wait for the world to respond, verify what happened, and tell you when it is done. Imagine asking it to coordinate a meeting with Rahul and getting a confirmed update back—not a list of instructions to follow yourself. The ambition is simple: make real work as easy to delegate as sending a message, without giving up the control needed to trust the result.

Built with TypeScript, Postgres, Vitest. Design constraints: Exactly-Once Actions and Bounded Authorization.

Software Engineering Intern at MetLife

MetLife, US Technology, May – Jun 2026

• TSTypeScript
• DBDatabases (Azure Cosmos DB)
• JVJava and Spring Boot
• PLPipelines and automation (Excel streaming job and schedulers)
• AILLMs and agents (Gen AI programming and multi-agent orchestration)
• QATesting (Playwright)

I rebuilt the backend behind an engineering-metrics dashboard, and it now runs in production.

Worked on Codenoesis, a Spring Boot engineering-metrics dashboard for Copilot usage, commits, sprint velocity, and story points.

• Rebuilt the legacy entity, job, and service layer into a maintainable Spring Boot backend pipeline now running in production.
• Designed four Azure Cosmos DB entities fed by a custom Excel streaming job with dedicated schedulers.
• Integrated Azure DevOps APIs for program-increment date refinement and migrated commit ingestion to GitHub Enterprise APIs.
• Applied TypeScript, Java, Spring Boot, Playwright, Gen AI programming, and multi-agent orchestration in production.

The Himalayan Trout House

Client website

• TSTypeScript (Next.js)

A website for a real client. When the owners need a change, they text me and I hard-code it.

The problem

The client needed a polished web presence with optimized media, an enquiry flow, and trustworthy social proof.

What I built

Built a Next.js React 19 site with Cloudinary image optimization, Resend enquiry emails, and Google Places review integration, deployed on Vercel. When the owners need a change, they text me and I hard-code it.

Result

Delivered a production client website that I keep up to date by hand as the owners' needs change.

Built with Next.js, TypeScript, Cloudinary, Resend, Vercel.

Automated Job-Search & Application Platform

Multi-agent automation

• DBDatabases (Supabase)
• PLPipelines and automation (n8n and Apify)
• AILLMs and agents (Claude API agents)

Comparing job posts by hand is slow and inconsistent. This 51-node pipeline reads, ranks, and drafts the applications.

The problem

Manual job search workflows are slow, repetitive, and hard to compare consistently across seniority, risk, fit, and application quality.

What I built

Built an n8n pipeline using Claude API, Apify, Supabase, and LaTeX to scrape postings, score them through five agent lenses plus risk analysis, and generate tailored resumes and cover letters.

Result

Converted a time-intensive search process into a ranked daily report with customized PDF application materials generated automatically.

Built with n8n, Claude API, Apify, Supabase, LaTeX, Automation.

AI Second Brain

Knowledge retrieval

• AILLMs and agents (OpenAI API and RAG)
• PYPython

Notes get less useful as they pile up. This lets you ask an Obsidian vault questions in plain language.

The problem

Personal notes become less useful as they grow unless they can be searched semantically and revisited through natural questions.

What I built

Built an assistant that parses markdown notes, embeds the knowledge base, retrieves relevant context, and answers through an OpenAI-powered RAG workflow.

Result

Turned a note vault into a conversational knowledge interface, showing practical LLM application design beyond a simple chat wrapper.

Built with Python, OpenAI API, RAG, Vector Retrieval, Markdown Parsing.

WebscrapeAI

Scraping pipeline

• PLPipelines and automation (BeautifulSoup scraping)
• AILLMs and agents (LangChain)
• PYPython

Web pages rarely agree on structure. This pipeline pulls the data out and gives it a shape you can reuse.

The problem

Useful web data often arrives in inconsistent page structures, making simple scraping brittle and downstream processing messy.

What I built

Built a Python scraping workflow with BeautifulSoup and LangChain to extract page data, clean it, summarize it, and prepare structured outputs for reuse.

Result

Demonstrates practical automation thinking across extraction, transformation, and LLM-assisted structuring.

Built with Python, LangChain, BeautifulSoup, Web Scraping, LLM Workflows.

Smart Stock

Forecasting product

• PYPython (Streamlit)
• QATesting

A forecast isn’t a decision. Smart Stock turns 7‑day and 28‑day demand forecasts into reorder logic.

The problem

Forecasting projects often stop at model outputs, while inventory teams need clear safety-stock and reorder decisions.

What I built

Built a multi-horizon forecasting app that benchmarks against baselines, validates outputs, and automates safety-stock and reorder calculations.

Result

Packages forecasting, business logic, and interface design into a usable end-to-end retail analytics product.

Built with Python, Streamlit, Forecasting, Inventory Logic, Testing.