shravanshidruk16/innomatics-research-labs-advanced-genai-internship-february-2026

1. Repository shape

This is a multi-project internship workspace, not a single application. Each top-level folder is a largely independent deliverable: FastAPI assignments, LangChain experiments, an NLP set of tasks, a RAG customer-support assistant, and a hackathon dataset project.

2. Primary boundary: project isolation

The repo is organized by topic and submission, so architecture is best drawn as several separate subsystems rather than one shared platform. Most projects have their own code, data, and artifacts, with limited evidence of cross-project dependencies.

3. RAG Major Project – Customer Support Assistant

This is the clearest end-to-end production-style system. The app/ package is split into distinct pipeline stages: ingestion, retrieval, routing, LLM orchestration, workflow graph, and human-in-the-loop escalation. The presence of vectorstore/ and a knowledge_base.pdf indicates a document-grounded assistant built on indexed knowledge.

4. RAG data flow

The likely flow is: PDF knowledge base → loader → chunker → embedder → persisted vector store (Chroma) → retriever → routing/workflow → LLM response. The hitl/escalation.py module suggests a fallback path where uncertain or sensitive queries are escalated instead of answered directly.

5. RAG architecture boundary

The main architectural split is between offline ingestion/indexing and online query handling. Ingestion creates the vector store artifacts under vectorstore/, while runtime code in main.py, graph/workflow.py, retrieval/retriever.py, and routing/router.py serves user requests and decides response strategy.

6. LangChain Assignment 3

This folder is a smaller, modular LLM pipeline focused on resume-job matching. The code is structured into chains/ and prompts/, with main.py orchestrating execution over input text files such as job_description.txt, strong_resume.txt, average_resume.txt, and weak_resume.txt.

7. LangChain data flow

The pipeline appears to follow a staged analysis pattern: extract information → match resume to job → score fit → explain result. The separate prompt files and chain files indicate a design where each LLM step is independently defined and then composed in the main script.

8. FastAPI major project

Final Major Project - FastAPI/Online_Course_Platform_FastAPI/ is a standalone web/API application for an online course platform. The core split is between main.py for application entry/API routing, models.py for domain/data structures, and frontend.py for UI or client-side integration. assets/ and the generated HTML file suggest a packaged demo or report alongside the app code.

9. FastAPI assignments

IN226039402_FASTAPI/Assignment_1 through Assignment_5 are separate learning exercises, each with its own main.py, HTML export, and images. These are best treated as independent FastAPI practice modules rather than parts of one shared system.

10. NLP and data projects

The NLP/ folder contains notebook-driven experiments plus task3.py, with datasets for Kindle and IMDb reviews. The Entrance Test - Hackathon/ folder contains notebooks and a generated CSV dataset for food delivery. These are analysis/modeling workspaces, not app backends.

11. Python programming notebooks

Python Programming/ is a sequence of weekly notebook exercises. This area is instructional rather than system-oriented, and it does not appear to feed the major projects directly.

12. Supporting artifacts

The repo includes generated deliverables, HTML exports, PDFs, images, datasets, and a Chroma persistence directory. These artifacts matter architecturally because they show the projects are documented, reproducible exercises with persisted retrieval state rather than purely transient notebook experiments.