Skip to content
View shahmanager's full-sized avatar

Highlights

  • Pro

Block or report shahmanager

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. 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
shahmanager/README.md

πŸ‘¨β€πŸ’» About Me

Hi! I am a developer and builder passionate about AI systems, agentic workflows, and hands-on systems programming. I enjoy building functional projects from first principles, learning how complex systems work under the hood, and connecting high level AI models with practical applications and physical hardware.

β–Έ πŸ”­ What I am working on: Voice conversational agents, multi agent coordination setups, and full stack tools.
β–Έ ⚑ Areas of exploration: Real time speech pipelines (Pipecat, Gemini Live S2S), edge computing with Cloudflare Workers, and embedded C++ firmware.
β–Έ πŸ”¬ Curiosities: How autonomous agents manage long term memory, privacy preserving distributed ML (Federated Learning), and writing clean, reliable code.
β–Έ πŸ’¬ Let's talk about: Python, TypeScript, React, AI experiments, hardware tinkering, and building things that solve real problems.


πŸ› οΈ Tech Stack & Tooling

Skill Icons

Languages Python, TypeScript, JavaScript, C++, C, SQL, Bash
AI & Frameworks PyTorch, Gemini API, LiteLLM, Pipecat Voice AI, Hugging Face
Web & Backend React, Node.js, FastAPI, Cloudflare Workers, Cloudflare D1, Tailwind CSS
Tools & Storage PostgreSQL, SQLite (FTS5), Redis, Docker, Git, GitHub Actions, Linux
Hardware & Systems Embedded C++, Firmware Experiments, Hardware Interfacing

πŸ”€ Architecture Pipeline: Real Time Voice Agent

flowchart LR
    A["Caller Audio Stream"] --> B["Pipecat Audio Pipeline"]
    B --> C["Gemini Live S2S Engine"]
    C --> D{"Grounding & Safety"}
    D -->|"Query Cited Passages"| E[("SQLite FTS5 Manuals")]
    D -->|"Safety Short Circuit"| F["Safety Circuit Breaker"]
    D -->|"Synthesize Response"| G["Sarvam AI Speech"]
    G --> H["Live Telephony Stream"]

    style A fill:#0d1117,stroke:#58a6ff,stroke-width:2px,color:#c9d1d9
    style B fill:#0d1117,stroke:#58a6ff,stroke-width:2px,color:#c9d1d9
    style C fill:#0d1117,stroke:#bc8cff,stroke-width:2px,color:#c9d1d9
    style D fill:#0d1117,stroke:#f0883e,stroke-width:2px,color:#c9d1d9
    style E fill:#0d1117,stroke:#3fb950,stroke-width:2px,color:#c9d1d9
    style F fill:#0d1117,stroke:#f85149,stroke-width:2px,color:#c9d1d9
    style G fill:#0d1117,stroke:#58a6ff,stroke-width:2px,color:#c9d1d9
    style H fill:#0d1117,stroke:#58a6ff,stroke-width:2px,color:#c9d1d9
Loading

πŸš€ Projects & Experiments

A voice agent exploring live speech to speech conversations with grounded knowledge retrieval.

  • Speech to Speech: Conversational audio in Indian English and Hinglish using Gemini Live S2S and Pipecat.
  • Grounded Answers: Uses SQLite FTS5 text search over product manuals to cite passages instead of guessing.
  • Safety Guardrails: Simple keyword interceptors that prioritize user safety before triggering model calls.
  • Telephony: Streams live audio over WebSockets with state checkpointing between turns.

An experimental two agent setup exploring cloud execution, laptop verification, and memory.

  • Separates background cloud task execution from local laptop evidence review.
  • Explores durable agent memory and structured messaging between collaborating agents.
  • Focuses on clear progress logging and verifiable completion criteria.

A lightweight Python utility exploring multi agent file leases and SHA based build verification.

  • Built in a single standard library Python file with zero third party dependencies.
  • Uses simple file leases to prevent race conditions during concurrent agent edits.
  • Computes SHA hashes to ensure clean context boundaries and reproducible build steps.

A practical simulation of privacy preserving machine learning across banking nodes using FedAvg.

  • Simulates distributed financial institutions training local models on private client data.
  • Aggregates model weights centrally via Federated Averaging without sharing raw records.
  • Explores the balance between predictive performance and data privacy.

Pinned Loading

  1. billmart_chatbot billmart_chatbot Public

    Python

  2. federated-credit-risk federated-credit-risk Public

    Federated Credit Risk Analysis using FedAvg - privacy-preserving ML across simulated bank clients

    Python

  3. FusionRestore FusionRestore Public

    Python