Skip to content
View bytebymanas's full-sized avatar
🎯
Focusing
🎯
Focusing
2 followers · 3 following

Highlights

  • Pro

Block or report bytebymanas

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.

Content in all repositories owned by your account will be closed.
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
bytebymanas/README.md

> about

I build intelligent systems that turn messy data into useful decisions — from deep learning models and NLP pipelines to RAG systems, vector search, local LLMs, and multi-agent workflows.

Currently pursuing CSE (AI/ML) at Chandigarh University and working as an ML Intern at FlyRank, where I help build models and insights across 30,000+ web pages.

Python PyTorch TensorFlow NLP RAG LangChain Vector Databases FastAPI Ollama LLMs

> stack

Languages
ML / DL
LLMs · RAG · Agents
Backend · Tools

> projects

🧠 Yukti AI

Socratic DSA mentor, fully local

Guides you to the answer with questions instead of handing it over.

  • 3-layer context pipeline: rolling summary + 4-turn window + current message
  • Cross-session memory via async behavioral snapshots in SQLite
  • Cloud GPU inference over ngrok with local Ollama fallback

FastAPI Ollama Qwen2.5 SQLite

📰 Sift AI

Agentic news curation pipeline

Reads 15–20+ articles a day and keeps only what’s worth your time.

  • Multi-agent CrewAI flow scoring substance, credibility and learning value
  • Local Qwen2.5 inference with retry/fallback orchestration
  • RSS + REST ingestion, deduplication, auto PDF digest over SMTP

CrewAI Ollama WeasyPrint

📉 Fake News Detector

Deep learning text classifier

Separates fake from legitimate news using sequence models.

  • RNN/LSTM with Word2Vec embeddings and Byte-Pair Encoding
  • Accuracy lifted from 88% to 94.6% through model tuning

TensorFlow LSTM Word2Vec NLP

😴 Driver Drowsiness Detection

Real-time edge vision, no GPU

Flags a drowsy driver live on plain CPU hardware.

  • 478 facial landmarks at 25 FPS
  • Eye Aspect Ratio classifier at 94–96% accuracy
  • Co-authored a research paper on the approach

MediaPipe OpenCV NumPy

> activity

Blue animated contribution snake

> connect

Popular repositories Loading

  1. CodeSync CodeSync Public

    CodeSync – A Browser extension for backing up accepted LeetCode and GFG solutions to GitHub.

    JavaScript 4 2

  2. Sift-AI Sift-AI Public

    Automated daily AI research engine powered by CrewAI & serverless Nvidia T4 GPUs.

    Jupyter Notebook 2

  3. RATEMYBITE RATEMYBITE Public

    Python 1 1

  4. NOVA NOVA Public

    Python

  5. binance-trading-bot binance-trading-bot Public

    Python-based CLI Trading Bot for Binance Futures Testnet

    Python

  6. bytebymanas bytebymanas Public