System Requirements Document

1. Introduction

This document defines the requirements for a personal portfolio website for Nishil Modi, a Computer Engineering graduate and AI/ML Engineer. The site is a self-owned, single-author web presence whose purpose is to present professional identity, technical expertise, work experience, projects, education, and certifications to visitors such as recruiters, hiring managers, and technical peers.

The site must be simple but attractive, and its visual theme must reflect the owner's professional domain: artificial intelligence and machine learning engineering.

All content for the site is derived from the owner's resume, including the professional summary, technical skills, professional experience, projects, education, certifications, contact details, and outbound professional profile links (LinkedIn, GitHub, HackerRank).

2. System Overview

The product is a static, single-author portfolio website delivered as a frontend-only web experience. It contains no user accounts, no administration surface, no backend services, and no data persistence layer. All content is authored and maintained by the site owner in the site source.

The site is organized as a set of navigable content sections:

• Hero / identity introduction
• Professional summary
• Technical skills, grouped by category
• Professional experience
• Projects
• Education
• Certifications
• Contact and outbound professional links

Visitors consume the site read-only. The only interactive behavior is in-page navigation, hover and scroll presentation states, and outbound links to external professional profiles.

3. Functional Requirements

Identity & Overview

FR-1 — Hero identity section As a recruiter visiting the site, I want a hero section that displays Nishil Modi's name, professional title as an AI/ML Engineer, and location so that I immediately know who the site represents and what discipline they work in.

FR-2 — Domain-aligned visual identity As the site owner, I want the site's theme to visually signal an AI/ML engineering identity so that the first impression matches my professional field.

FR-3 — Professional summary display As a hiring manager, I want to read a concise professional summary describing the owner as a Computer Engineering graduate with specialized expertise in machine learning, NLP, and generative AI, currently prototyping autonomous AI Agent workflows as an AI/ML intern, with proven ability to architect and deploy end-to-end AI applications including LLM-powered RAG chatbots and recommendation engines.

FR-4 — Credential backing in summary As a hiring manager, I want the summary to state that the owner's experience is backed by certifications from DeepLearning.ai and Oracle and that the owner works hands-on with Python, LangChain, and vector databases so that I can assess credibility and tooling fit.

FR-5 — Stated career objective As a recruiter, I want the site to convey that the owner is seeking to apply hands-on experience to develop and ship production-ready AI solutions in a full-time role, so that I understand availability and intent.

Technical Skills

FR-6 — Skills section grouped by category As a technical reviewer, I want the technical skills presented as discrete, labeled categories so that I can scan capabilities quickly rather than reading a flat list.

FR-7 — Programming languages listing As a technical reviewer, I want the Languages category to list Python and SQL.

FR-8 — AI/ML techniques listing As a technical reviewer, I want the AI/ML Techniques category to list Machine Learning, Deep Learning, NLP, LLM, RAG, Generative AI, AI Agent, CNN, RNN, LSTM, Supervised Learning, Unsupervised Learning, Recommender Systems, and Tool Calling.

FR-9 — AI/ML frameworks listing As a technical reviewer, I want the AI/ML Frameworks category to list Scikit-learn, LangChain, LangGraph, Pandas, NumPy, Matplotlib, Seaborn, FastAPI, and Hugging Face.

FR-10 — Frontend skills listing As a technical reviewer, I want the Frontend category to list HTML, CSS, Material UI, Tailwind CSS, and Bootstrap.

FR-11 — Databases and tools listing As a technical reviewer, I want the Databases & Tools category to list PostgreSQL, FAISS, Vector Databases, and Semantic Search.

FR-12 — Developer tools listing As a technical reviewer, I want the Developer Tools category to list VS Code, Google Colab, and GitHub.

FR-13 — Core concepts listing As a technical reviewer, I want the Core Concepts category to list Data Structures and Algorithms (DSA), Database Management Systems (DBMS), Object-Oriented Programming (OOP), and Exploratory Data Analysis (EDA).

Professional Experience

FR-14 — Professional experience section As a hiring manager, I want a professional experience section that lists roles with employer, role title, and date range so that I can evaluate career progression.

FR-15 — AI/ML Intern role at Neuronet Systems As a hiring manager, I want to see the AI/ML Intern role at Neuronet Systems dated May 2026 – August 2026.

FR-16 — Autonomous agent workflow responsibility As a hiring manager, I want the Neuronet Systems role to describe architecting and prototyping AI Agent workflows using Python and evaluating and benchmarking frameworks for building autonomous, multi-agent systems.

FR-17 — LLM agent architecture responsibility As a hiring manager, I want the Neuronet Systems role to describe engineering foundational architectures for LLM-powered agents supporting dynamic tool calling, API integration, and complex multi-step task execution.

Projects

FR-18 — Projects section As a technical visitor, I want a projects section where each project is presented as a distinct entry with its name, date, and detail bullets so that I can assess depth of hands-on work.

FR-19 — PriceHunt project entry As a technical visitor, I want a project entry for PriceHunt (AI Agent), dated June 2026.

FR-20 — PriceHunt full-stack price comparison capability As a technical visitor, I want the PriceHunt entry to describe a full-stack AI-powered price comparison platform for India's fragmented e-commerce market built with FastAPI and React, aggregating real-time prices from Amazon, Flipkart, Myntra, and Ajio through a single natural language or Hinglish search query.

FR-21 — PriceHunt LangGraph agent pipeline capability As a technical visitor, I want the PriceHunt entry to describe a 6-node LangGraph AI agent pipeline covering intent parsing, search, aggregation, filtering, formatting, and error handling, powered by Google Gemini 2.5 Flash for query understanding and intent extraction.

FR-22 — PriceHunt product-grouping and lowest-price capability As a technical visitor, I want the PriceHunt entry to describe a smart product-grouping algorithm that identifies identical products across platforms and surfaces the lowest price with direct redirect links, backed by SerpApi Google Shopping for real-time data.

FR-23 — Chat with PDF Chatbot project entry As a technical visitor, I want a project entry for Chat with PDF Chatbot (Retrieval-Augmented Generation), dated February 2026.

FR-24 — Chat with PDF full-stack RAG capability As a technical visitor, I want the Chat with PDF entry to describe a full-stack RAG application built with Python and Streamlit that enables PDF ingestion and context-aware information extraction through a conversational, real-time streaming interface.

FR-25 — Chat with PDF document pipeline capability As a technical visitor, I want the Chat with PDF entry to describe a document pipeline engineered with LangChain and HuggingFace embeddings (all-MiniLM-L6-v2) using a FAISS vector database for accurate semantic retrieval.

FR-26 — Chat with PDF grounded-response capability As a technical visitor, I want the Chat with PDF entry to describe integration of Llama-3 (70B) via the Groq API with conversational memory and strict system prompts to prevent hallucinations, ensuring factually grounded, context-aware responses.

FR-27 — Movie Recommendation System project entry As a technical visitor, I want a project entry for Movie Recommendation System (Machine Learning), dated October 2025.

FR-28 — Movie recommendation web application capability As a technical visitor, I want the Movie Recommendation System entry to describe a machine learning web application built with Python and Streamlit that generates personalized movie recommendations through an interactive user interface.

FR-29 — Content-based filtering capability As a technical visitor, I want the Movie Recommendation System entry to describe a content-based filtering pipeline implemented using Scikit-Learn that vectorizes movie metadata and computes Cosine Similarity matrices to identify and rank related films.

FR-30 — NLP preprocessing capability As a technical visitor, I want the Movie Recommendation System entry to describe processing movie datasets with NLP text-preprocessing that consolidates genres, cast, and plot summaries into unified feature tags.

Education

FR-31 — Education section As a hiring manager, I want an education section listing the institution, location, date range, degree, and academic result.

FR-32 — Degree and institution entry As a hiring manager, I want to see a B.Tech in Computer Engineering from Leelaben Dashrathbhai Ramdas Patel Institute of Technology and Research, Gandhinagar, Gujarat, for the years 2022–2026, with a CGPA of 8.05/10.

Certifications

FR-33 — Certifications section As a hiring manager, I want a certifications section listing each certification with its title and issuing body so that I can verify external validation of skills.

FR-34 — Machine Learning Professional certification As a hiring manager, I want the certifications to include Machine Learning Professional – deeplearning.ai × Stanford University Online (Coursera).

FR-35 — Oracle Generative AI certification As a hiring manager, I want the certifications to include Oracle Cloud Infrastructure Generative AI Certified Professional – Oracle.

FR-36 — Python for Data Science certification As a hiring manager, I want the certifications to include Python for Data Science – IIT Madras (NPTEL).

FR-37 — Front-End certification entries As a hiring manager, I want the certifications to include Introduction to Front-End – META (Coursera) and HTML & CSS in depth – META (Coursera), listed as separate entries.

Contact & Outbound Links

FR-38 — Contact section As a recruiter, I want a contact section providing the owner's email and phone number so that I can initiate outreach directly.

FR-39 — Email contact action As a recruiter, I want the email address to be an actionable mail link that opens my mail client.

FR-40 — Phone contact action As a recruiter, I want the phone number to be an actionable call or messaging link.

FR-41 — LinkedIn profile link As a recruiter, I want a working hyperlink to the owner's LinkedIn profile.

FR-42 — GitHub profile link As a recruiter, I want a working hyperlink to the owner's GitHub profile.

FR-43 — HackerRank profile link As a technical reviewer, I want a working hyperlink to the owner's HackerRank profile.

FR-44 — Outbound link behavior As a visitor, I want external profile links to open in a new tab so that I do not lose my place on the portfolio.

Navigation & Presentation

FR-45 — In-page navigation As a visitor, I want a navigation affordance that lets me jump directly to any content section (summary, skills, experience, projects, education, certifications, contact).

FR-46 — Active section indication As a visitor, I want the navigation to indicate which section I am currently viewing.

FR-47 — Responsive layout As a visitor on a mobile device, I want the layout to reflow cleanly across phone, tablet, and desktop widths so that all content remains readable and all links remain tappable.

FR-48 — Scannable content hierarchy As a recruiter skimming the site, I want strong visual hierarchy between section headings, entry titles, metadata (dates, employers, institutions), and detail bullets so that I can extract key facts in a short scan.

FR-49 — Chronological ordering As a hiring manager, I want experience, projects, and education entries ordered most-recent-first.

FR-50 — Simple but attractive presentation As the site owner, I want a clean, uncluttered visual presentation with restrained decoration so that the site reads as professional rather than busy.

4. User Personas

Persona 1 — Recruiter / Hiring Manager Active visitor with the primary evaluation workflow. Scans identity, title, summary, skills, experience, and certifications; assesses seniority and role fit; then uses the contact section or LinkedIn link to initiate outreach. Cares about dates, employer names, role titles, and stated objective.

Persona 2 — Technical Peer / Engineering Evaluator Active visitor with a depth-oriented workflow. Moves directly to the projects section, reads each project's architecture and implementation bullets, and follows GitHub or HackerRank links to inspect code and problem-solving activity. Cares about named tools, frameworks, and system design details.

Persona 3 — Site Owner / Author Maintains the site content in the source and publishes updates. The owner is the sole author of all content and the only actor who changes the site. This persona does not introduce authentication, administration, or a content management surface.

System Actors

• External professional platforms (LinkedIn, GitHub, HackerRank) — outbound link targets only.
• Static hosting provider — serves the built site assets to visitors.

5. Core User Flows

Flow A — Recruiter quick qualification

1. Visitor lands on the hero section and reads name, AI/ML Engineer title, and location.
2. Visitor reads the professional summary and stated objective.
3. Visitor scans the skills section by category.
4. Visitor reads the professional experience entry (Neuronet Systems, AI/ML Intern, May 2026 – August 2026) and its responsibility bullets.
5. Visitor skims formatted project entries and certifications.
6. Visitor opens the contact section and uses the email link, phone link, or LinkedIn profile link to reach out.

Flow B — Technical peer project deep dive

1. Visitor lands on the hero section.
2. Visitor uses in-page navigation to jump directly to Projects.
3. Visitor reads the PriceHunt entry, including the 6-node LangGraph pipeline, Gemini 2.5 Flash usage, and SerpApi-backed aggregation details.
4. Visitor reads the Chat with PDF Chatbot entry, including the LangChain + HuggingFace + FAISS pipeline and Llama-3 70B via Groq.
5. Visitor reads the Movie Recommendation System entry, including Scikit-Learn content-based filtering and cosine similarity.
6. Visitor follows the GitHub link in a new tab to inspect the owner's code.

Flow C — Credential verification

1. Visitor opens the Education section and confirms institution, location, degree, years, and CGPA.
2. Visitor opens the Certifications section and reads each certification with its issuing body.
3. Visitor follows the LinkedIn or HackerRank link to cross-check the profile.

Flow D — Mobile scan

1. Visitor opens the site on a phone browser.
2. Hero, summary, skills, experience, and projects reflow into a single readable column.
3. Visitor navigates between sections using the responsive navigation affordance.
4. Visitor taps the email, phone, LinkedIn, or GitHub link directly from the device.

Flow E — Owner content update

1. Owner edits the relevant content section in the site source.
2. Owner adds or updates a project, certification, or experience entry.
3. Owner rebuilds and deploys the static site.
4. Updated content becomes visible to visitors without any backend or database change.

6. Visuals, Colors and Theme

The theme must read as AI/ML engineering while remaining simple and attractive. The authoritative creative direction for this project is a dual-nature craft treatment — an AI engineer shown as half mind, half machine — built on a hard 50/50 split between a warm paper ground and a deep ink ground, with one confident burnt-orange accent. The palette, typography, shape language, layout, motion, imagery, and hero direction below are binding; the generic indigo/blue-on-white SaaS template is forbidden for this project.

6.1 Palette (light mode)

• Background: Warm paper white #FBFAF7 as the page ground — deliberately not sterile #FFFFFF.
• Surface: #FFFFFF for cards and the code panel only.
• Text / primary: Ink #111214 carries all headings and the dark half of the split.
• Accent: Burnt signal orange #E4572E, used ONLY for interactive state: link underlines, the active nav marker, the cursor highlight, and the accent bar on the current role.
• Muted: #6E7076 for metadata, dates, and secondary labels.
• Proportion: Roughly 70% paper, 20% ink (the dark code panel and footer), 10% accent.
• Color contrast: All text and interactive elements maintain accessible contrast against their backgrounds.

6.2 Typography

• Headings: Space Grotesk — 700 for the name and section titles, tight tracking (-0.02em), sentence case for section headers but ALL-CAPS for micro-labels with 0.12em tracking.
• Name scale: The name is set enormous — 96px on desktop, 56px on mobile — as the single largest element on the page.
• Body: Instrument Sans.
• Scale: 1.333 modular scale — 96 / 56 / 40 / 28 / 20 / 16 / 14. Body 16px at 1.65 line-height.
• Monospace: JetBrains Mono at 13px for code snippets, tech tags, and data labels.
• Forbidden faces: No Inter, Roboto, Arial, Helvetica, Poppins, or system-ui for any type.

6.3 Shape language

• Sharp corners everywhere — 0px radius on cards, buttons, and image frames.
• The only exceptions are 2px on the tag chips and a 1px hairline border on everything.
• No soft blobs, no pill buttons, no rounded cards. The rigidity is the point: it reads as engineering, not marketing.

6.4 Layout

• Split-screen foundation: The hero is a 50/50 vertical split — left half warm paper with the name, title, and location in ink; right half a deep ink panel (#111214) containing a live-feeling terminal/code block rendering the owner's stack.
• Reading column: Below the hero, content flows in a single 720px reading column with a persistent left rail for section navigation.
• Project rows: Each project is a two-column row — metadata and dates in a narrow left column, description bullets in the wide right column — separated by 1px hairline rules, never cards.

6.5 Motion

• Restrained tempo. One purposeful reveal: on scroll, project rows fade up 12px with a 400ms cubic-bezier(0.16,1,0.3,1) ease, staggered 60ms.
• The hero's code panel types its first line once on load.
• Hover on any project row slides a 2px accent bar in from the left and shifts the row 4px right.
• No parallax, no particles, no bouncy springs.

6.6 Imagery

• No stock photography, no 3D blobs, no illustrations of people.
• The "imagery" is the interface itself: code blocks showing real LangGraph node definitions, a small architecture diagram of the 6-node PriceHunt pipeline drawn in 1px ink lines, technology tag chips, and the owner's initials "NM" set in Space Grotesk as the only logo mark.
• A single monochrome portrait is permitted in the About section, cropped square and treated with a subtle ink duotone.

6.7 Hero direction

• A full-viewport 50/50 split. Left: warm paper #FBFAF7, with "Nishil Modi" set in Space Grotesk 700 at 96px spanning nearly the full half-width, stacked on three lines ("Nishil" / "Modi" / then "AI/ML Engineer" in 28px Instrument Sans with a 2px accent underline). Below the name, a one-line location and a row of three text links — LinkedIn, GitHub, HackerRank — each with a 1px ink underline that fills accent on hover.
• Right: a solid #111214 panel, edge-to-edge, containing a terminal window with a warm-grey top bar and three dots, inside which a monospace block shows a real LangGraph agent snippet, with the accent color used only for the cursor and the agent.invoke() call.
• The split is hard-edged and full-bleed — no centered headline, no gradient, no blue button.

6.8 Avoid

• No blue, indigo, or violet anywhere — the accent is burnt orange #E4572E only.
• No gradient blobs, no glassmorphism, no frosted panels.
• No grid of identical hover-lift cards — projects are ruled rows, not tiles.
• No rounded pill buttons or soft card radii — corners stay sharp.
• No stock photography of people at laptops or abstract AI brain imagery.
• No centered hero with headline + subtext + blue CTA.
• No bouncy or springy motion — easing is deliberate and quiet.

7. Signature Design Concept

Concept: "Dual-Nature Craft" — half mind, half machine

The site's distinctive visual idea is the hard 50/50 split: a warm paper half carrying the oversized name, and a deep ink half carrying a live code panel — the visual thesis of an engineer who designs. The muse is Adham Dannaway, the designer/developer split, which maps exactly onto Nishil as a hybrid who architects AI systems (LangGraph 6-node pipelines, FAISS vector stores) and also builds the frontend (React, Tailwind). Crisp white with a deep ink ground and one confident accent keeps the site "simple but attractive," and the monospace voice gives the technical half its own honest register without falling into the blue-on-white SaaS template.

Signature moves

• A hard 50/50 split hero: warm paper half with the oversized name, deep ink half with a live code panel.
• Project entries as hairline-ruled rows (not cards): narrow left column for date and role, wide right column for bullets, with a 2px accent bar that slides in on hover.
• A persistent left rail of section markers (Work, Skills, About, Contact) rendered as small monospace labels with a single accent tick indicating the active section on scroll.
• The name "Nishil Modi" set at 96px in Space Grotesk, stacked across three lines, with the title "AI/ML Engineer" underlined in accent directly beneath — a typographic gesture that owns the first screen.
• A skills section presented as a monospace table with tabular alignment: category label in the left column, comma-separated values in the right, each row separated by a 1px hairline — reads like a config file, not a tag cloud.

The overall impression: technically credible, quietly confident, and uncluttered — simple in structure, distinctive in treatment. The emotional register is credible-but-warm: a builder who ships end-to-end systems, not a researcher; a working engineer's notebook that someone bothered to design beautifully — precise, technical, and human at the same time.

8. Interaction Model & Motion Direction

Interaction model

• Single-page layout with section anchors and a persistent navigation affordance.
• Navigation highlights the section currently in view as the visitor scrolls.
• Clicking a navigation item smooth-scrolls to the target section.
• Experience, project, education, and certification rows signal interactivity on hover: the row shifts 4px right and a 2px accent bar slides in from the left.
• Skill tags and project stack tags have a subtle hover state.
• All outbound profile and contact links open in a new tab.

Motion direction

• Entrance motion: content fades in and rises a short distance as it enters the viewport, on a short duration with a gentle ease-out.
• Scroll transitions are staggered slightly across sibling items (skills, project bullets, certification rows) to create a light cascade rather than a uniform pop.
• The hero's code panel types its first line once on load; the hero's ink panel and paper half remain otherwise still.
• No parallax, no auto-playing carousels, no looping attention-grabbing animation.
• All motion respects the visitor's reduced-motion preference and degrades to static presentation when that preference is set.

Landing Hero Motion Brief

• Tempo: Restrained (per the creative direction's Motion Tempo: restrained), with a bold hero drama register.
• Thesis (input → transformation → outcome): The visitor's arrival (input) is answered by a single typed line of a real LangGraph agent snippet appearing in the ink panel (transformation), leaving the visitor with the impression of a working engineer's live notebook (outcome).
• Focal subject: The monospace code block inside the terminal window on the ink half of the split.
• Visible layers: (1) warm paper #FBFAF7 left half with the oversized stacked name and accent-underlined title; (2) solid #111214 right half, edge-to-edge; (3) terminal window with warm-grey top bar and three dots; (4) monospace code block; (5) accent-colored cursor and agent.invoke() call.
• Loop: One contained, non-repeating sequence — the first code line types once on load, then the panel holds static. No looping animation.
• Composed first frame: The full 50/50 split already composed at rest: name fully set on the paper half, terminal window framed on the ink half, code block empty except the blinking accent cursor.
• Optional interaction: Hovering the three text links (LinkedIn, GitHub, HackerRank) fills their 1px ink underline with accent; no other hero interaction is required.
• Responsive behavior: On mobile the split stacks — paper half with the name at 56px first, ink code panel beneath, both full-bleed and hard-edged.
• Reduced-motion fallback: The typing sequence is suppressed; the first code line renders fully formed and static, and the cursor does not blink.

Landing Hero 3D Scene Brief — DIRECTION-DERIVED

Not applicable. The creative direction specifies hero dimensionality: layered_2d, so the hero is composed from 2D DOM/SVG/CSS layers only. No Canvas, WebGL, Three.js, or R3F scene is required, and no 3D wording is implied by this brief.

9. Non-Functional Requirements

• NFR-1 Performance: The site must load quickly on typical mobile connections; assets should be lightweight and images optimized.
• NFR-2 Static delivery: The product is frontend-only. No server-side runtime, no database, and no API is required to render content.
• NFR-3 Responsiveness: Layouts must render correctly at mobile, tablet, and desktop breakpoints, with no horizontal overflow.
• NFR-4 Accessibility: Semantic HTML landmarks and heading order, keyboard-operable navigation and links, visible focus states, sufficient color contrast, and non-text-content alternatives for decorative imagery.
• NFR-5 Cross-browser compatibility: Consistent rendering in current versions of Chrome, Firefox, Safari, and Edge.
• NFR-6 Content accuracy: All displayed resume data (dates, titles, employers, CGPA, metrics, tool names) must match the owner's source resume exactly.
• NFR-7 Content maintainability: Content must be editable in the site source in a structured way so the owner can add or update a project, certification, or experience entry without restructuring the layout.
• NFR-8 Link integrity: All outbound links (LinkedIn, GitHub, HackerRank) and contact links must resolve correctly and be verified before publication.
• NFR-9 Search discoverability: Page title, meta description, and basic structured metadata must reflect the owner's name and AI/ML Engineer identity.
• NFR-10 Visual consistency: Accent colors, spacing scale, typography scale, and surface treatment must be applied consistently across every section, and the palette, typefaces, shape language, and layout defined in Section 6 must not be substituted with a generic template.
• NFR-11 Reduced motion: Animation must be suppressible through the operating system's reduced-motion setting.

10. Tech Stack

Derived from the accepted delivery shape (a static, single-author portfolio) and the technologies the owner has stated proficiency in:

• Markup & styling: HTML and CSS.
• Styling framework: Tailwind CSS for layout, theming, and responsive breakpoints; Material UI and Bootstrap are listed as owner skills but are not required by this static site's scope.
• Scripting: Vanilla JavaScript for in-page navigation, active-section highlighting, scroll-reveal behavior, and reduced-motion handling.
• Build approach: Static site build producing plain HTML/CSS/JS assets.
• Hosting: Static hosting provider (e.g., GitHub Pages or an equivalent static host) serving the built assets directly.
• No application layers required: No backend framework, no database, no authentication/identity layer, no integration service, and no content management system are part of this product.

11. Assumptions and Constraints

• The site is a single-author portfolio owned and maintained by Nishil Modi; there are no other content contributors.
• The visual direction is fixed to the dual-nature craft treatment in Section 6: warm paper #FBFAF7 ground, ink #111214 dark half, burnt orange #E4572E as the only accent, Space Grotesk / Instrument Sans / JetBrains Mono, sharp 0px corners, and a hard 50/50 split hero. No blue, indigo, or violet may be introduced.
• All content is sourced from the owner's resume; no additional personal, private, or unlisted information is published.
• The site is delivered as a single-page layout with section anchors.
• The product is frontend-only and static; there is no server, database, or user account state.
• Contact is one-directional — visitors reach the owner through the published email, phone, or LinkedIn; no contact form or messaging system is in scope.
• LinkedIn, GitHub, and HackerRank are external destinations; the site links out to them and does not embed or duplicate their data.
• The design must remain "simple but attractive": restraint in decoration takes priority over feature richness.
• Project and experience dates as stated in the resume (May 2026 – August 2026, June 2026, February 2026, October 2025, 2022–2026) are reproduced as given.
• Content updates are performed by editing the site source and redeploying.

12. Glossary

• AI Agent: An autonomous software system that plans and executes multi-step tasks, including dynamic tool calling and API integration.
• CNN: Convolutional Neural Network, a deep learning architecture commonly used for grid-like data such as images.
• Cosine Similarity: A similarity measure between vectors used to rank related items in recommendation systems.
• DBMS: Database Management Systems.
• DSA: Data Structures and Algorithms.
• EDA: Exploratory Data Analysis.
• Embeddings: Dense vector representations of text used for semantic comparison and retrieval.
• FAISS: A library for efficient similarity search over dense vector embeddings.
• FastAPI: A Python web framework used to build the backend of the PriceHunt platform.
• Gemini 2.5 Flash: The Google model used in PriceHunt for query understanding and intent extraction.
• Generative AI: AI systems that produce new content such as text or code.
• Groq API: The inference API used to serve Llama-3 (70B) in the Chat with PDF chatbot.
• Hinglish: Code-mixed Hindi–English text, supported as a natural-language search input in PriceHunt.
• Hugging Face: Provider of models and tooling, including the all-MiniLM-L6-v2 embedding model used in the RAG pipeline.
• LangChain: A framework used to build LLM-powered applications and document pipelines.
• LangGraph: A framework for building stateful, multi-node agent workflows, used for PriceHunt's 6-node pipeline.
• Llama-3 (70B): The large language model used for grounded conversational responses in the Chat with PDF chatbot.
• LLM: Large Language Model.
• LSTM / RNN: Recurrent neural network architectures for sequential data.
• NLP: Natural Language Processing.
• Node (agent pipeline): A discrete processing stage within an agent workflow; PriceHunt has six: intent parsing, search, aggregation, filtering, formatting, and error handling.
• OOP: Object-Oriented Programming.
• RAG: Retrieval-Augmented Generation — combining semantic document retrieval with LLM generation to produce grounded answers.
• Recommender System: A system that ranks and suggests items to a user, here implemented with content-based filtering.
• SerpApi Google Shopping: The real-time product data source backing PriceHunt's price aggregation.
• Semantic Search: Retrieval based on meaning similarity rather than literal keyword matching.
• Streamlit: The Python framework used to build the interactive interfaces for the Chat with PDF chatbot and the Movie Recommendation System.
• Tool Calling: The ability of an LLM agent to invoke external functions or APIs during task execution.
• Vector Database: A data store optimized for storing and querying embeddings; FAISS is used in this portfolio's projects.