Skip to content
View 45Harry's full-sized avatar
🦇
🦇
🦇
🦇
24 followers · 28 following

Achievements

Achievement: Pair ExtraordinaireAchievement: YOLOAchievement: QuickdrawAchievement: Pull Sharkx2

Achievements

Achievement: Pair ExtraordinaireAchievement: YOLOAchievement: QuickdrawAchievement: Pull Sharkx2

Block or report 45Harry

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
45harry/README.md

Hi 👋, I'm Hari Narayan Sah

AI Engineer · Production Voice Agents · Speech · Computer Vision

Coding

45harry

  • 🔭 AI Engineer at Mantra Ideas, Lalitpur, shipping production voice systems and clinical speech pipelines
  • 🎙️ Built Colait's physician interview agent and its clinical scribe pipeline
  • 🌉 Built AgentBridge solo: move AI coding sessions between Claude Code, Codex, OpenCode and Antigravity
  • 🇳🇵 Working on Nepali TTS: Chatterbox and Matcha-TTS fine-tunes for Devanagari
  • 💬 Ask me about voice agents, LiveKit, Bedrock, speech pipelines, and computer vision
  • 🌐 Portfolio: harins.com.np
  • 🤗 Hugging Face: Imbatmann
  • 📫 How to reach me: smartharry499@gmail.com · LinkedIn

Featured Work:

  • AgentBridge: my solo open-source project. One session layer for every AI coding agent: start a conversation in Claude Code and continue it in Codex, OpenCode or Antigravity, from any folder. A small Rust MCP server that reads sessions straight from the original files and writes continued turns back. pipx install agentbridge-rs
  • Colait: clinical voice interview and scribe platform. Physician interviews over bidirectional audio or chat. Scribe turns a visit into a diarized, romanized brief. Deployed to ECR and Bedrock AgentCore. (private)
  • TelvoxAI: LiveKit voice-agent platform with a multi-tool phone agent (leads, calendar, transfer, custom HTTP tools). Moving STT to Groq cut turn latency from ~8–9s to ~1–2s. (private)
  • Nepali TTS: chatterbox-nepali-tts on Hugging Face
  • BanknoteOCR: Nepali denomination and signature pipeline (Roboflow detector + PyTorch siamese signature model, FastAPI), ~90% production accuracy.
  • AnswerServiceAI: UK hotel order and garage booking voice agents on Vapi, n8n, and Twilio.

Languages and Tools:

python aws azure fastapi docker pytorch opencv huggingface langchain livekit c cplusplus git linux

45harry

45harry

Pinned Loading

  1. FooDBOT_END_TO_END_IN_DIALOGFLOW_NLP FooDBOT_END_TO_END_IN_DIALOGFLOW_NLP Public

    A Simple FoodBot for a IndoNep Eatery :- Saavita- The FoodBot

    HTML

  2. Customer_Churn_predictor_using_ANN Customer_Churn_predictor_using_ANN Public

    The Churn predictor model designed using Neural Net . It takes age,salary etc as input and predicts weather the person will leave or not

    Jupyter Notebook

  3. Potato_Disease_Classification Potato_Disease_Classification Public

    Potato Disease Classification - Traning, Rest Api and FrontEnd to Test

    Jupyter Notebook

  4. Cat_or_Dog_classification_using_CNN Cat_or_Dog_classification_using_CNN Public

    Cat or Dog Binary Classifier using CNN. This project is a Convolutional Neural Network (CNN) built with TensorFlow and Keras to classify images of cats and dogs.

    Jupyter Notebook

  5. Face_Recognition_Attendance_project Face_Recognition_Attendance_project Public

    Face Attendance System - OpenCV, Face-Recgnition

    Python

  6. Log_Classification_GenAI_LLM_NLP_DeepSeek_R1_ Log_Classification_GenAI_LLM_NLP_DeepSeek_R1_ Public

    Log Classification Using LLM DeepSeekR1

    Jupyter Notebook