Skip to content
View afianas's full-sized avatar
47 followers · 49 following

Achievements

Achievement: YOLOAchievement: Pull Shark

Achievements

Achievement: YOLOAchievement: Pull Shark

Block or report afianas

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
afianas/README.md

Afia Nasumudeen

About Me

I'm a Computer Science and Business Systems student working on solving real world problems.

  • Currently architecting: High-precision RAG pipelines for meeting intelligence.
  • Exploring: LLM agents, multi-modal AI, and scalable backend architectures
  • Interested in: LLMs & RAG · Computer Vision · ML

Featured Projects

📊 Meetric — Meeting Intelligence Hub

RAG platform extracting structured decisions from raw conversations.

  • Engineered a dual-stage retrieval architecture combining Pinecone (Serverless) + BGE-Reranker (Cross-Encoder) with Llama 3.3-70B for high-precision, grounded extraction and synthesis.
  • ⚙️ LLM Integration: Architected a Llama 3.3-70B powered extraction engine using advanced Prompt Engineering to convert unstructured transcripts into action items and decisions.
  • 🔍 Explainability: Full traceability linking every decision back to exact transcript segments.
  • 🛠️ Tech: FastAPI, React, TypeScript, Pinecone, Groq, DistilBERT.

🧠 SpellCheck - AI Spelling Tutor

Hybrid NLP system for contextual error detection and adaptive learning.

  • ⚙️ Hybrid Inference: Combining Character-level N-grams with a fine-tuned DistilBERT transformer.
  • 🎯 Error Classification: Developed a multi-class classifier to identify specific student error types (Vowel Confusion, Letter Swaps etc.) for targeted feedback.
  • 🛠️ Tech: Python, Transformers, Expo (Mobile), gTTs.

🍽️ BiteSmart — Food Recognition & Calorie Estimator

Computer Vision system for real-time dietary intelligence.

  • Leveraged EfficientNetB3 with fine-tuning, achieving 82.8% Top-1 accuracy and 94.5% Top-5 accuracy on Food-101.
  • 📈 Impact: Instant calorie estimation via fuzzy matching (Fuse.js) against a local nutritional database.
  • 🛠️ Tech: TensorFlow, Keras, FastAPI, React 19 (Vite), Framer Motion.

🔊 Invora — Voice Commerce for MSMEs

Multilingual, voice-first inventory intelligence for small-scale suppliers.

  • Integrated OpenAI Whisper ASR with a custom rule-based NLP parser for Malayalam, Hindi, and English.
  • 📈 Impact: Warehouse management with Pareto-based (ABC) revenue analysis for shopkeepers.
  • 🛠️ Tech: React Native (Expo), FastAPI, Whisper, RapidFuzz.

Tech Stack

Languages

Python JavaScript TypeScript HTML5 CSS3

AI / ML

TensorFlow Keras NumPy Pandas scikit--learn HuggingFace

Backend & Databases

FastAPI MongoDB SQLite

Frontend & Tools

React Expo Git Postman

GitHub Stats

Let's Connect

I'm open to collaborations, internships, and conversations about building the next generation of intelligent systems.

"The best way to predict the future is to engineer it."

Pinned Loading

  1. Meetric-Meeting-Intelligence-Hub Meetric-Meeting-Intelligence-Hub Public

    End-to-end system for transcript analysis using multi-tier LLM ingestion, Pinecone vector search, and reranking to deliver verifiable insights, traceable decisions, and contextual Q&A.

    TypeScript 2 1

  2. SpellCheck SpellCheck Public

    Hybrid AI spelling tutor combining rule-based logic, character-level ML, and a fine-tuned DistilBERT model to detect spelling errors and deliver educational feedback.

    Jupyter Notebook

  3. BiteSmart BiteSmart Public

    BiteSmart is a deep learning–based food recognition system on the Food-101 dataset that classifies food images, estimates calories, and recommends healthier alternatives using computer vision.

    Jupyter Notebook

  4. Invora_Voice_Commerce Invora_Voice_Commerce Public

    Invora Voice Commerce is a multilingual voice-first inventory management system that converts spoken supplier transactions into structured inventory updates using speech recognition and rule-based …

    JavaScript

  5. DermaScan DermaScan Public

    DermaScan is a mobile-ready deep learning application that classifies acne severity into Mild, Moderate, and Severe using a lightweight CNN model (MobileNetV2).

    Jupyter Notebook

  6. Employee-Attrition-Prediction Employee-Attrition-Prediction Public

    End-to-end ML project for employee attrition prediction using Random Forest, SMOTE, and Streamlit deployment.

    Jupyter Notebook