Muhammad H.

I build AI agents, RAG chatbots and voice agents that real users rely on every day, plus the Python and React products around them. 8+ years in full-stack engineering, Top Rated Plus, and platforms I've worked on serve 30,000+ users.

RECENT PROJECTS

White-label AI agent platform (Stammer AI, 30,000+ users, thousands of MRR)
Businesses build and resell their own chat and voice agents. We split each customer's content into small chunks, give every agent its own Pinecone namespace, and only answer above a minimum match score, so the bot hands off instead of guessing. Voice agents run on Retell AI and Twilio. [Django, React, OpenAI, Claude, Pinecone, Stripe]

AI sales call coaching (Gold Selling Standards)
Zoom and Teams calls are scored on 12 sections from 0 to 10 with an evidence-only rule, so every score points to the exact lines in the transcript. Each call type (Discovery, Demo, Renewal) has its own weighted rubric, and a chatbot answers questions about a deal's past calls. [OpenAI, Pydantic, Pinecone, Microsoft Graph]

Meetings to CRM, automatically (SalesEdgeAI)
Finished Zoom and Teams recordings are pulled in with no bot joining the call, analyzed, and pushed to Salesforce, HubSpot or Zoho: opportunity, transcript, insights, follow-up tasks and a draft email. [Django, AWS SQS, n8n]

AI lead filter for social listening (Notifier, 30,000+ users)
Mentions from Reddit, X, LinkedIn, YouTube and more are checked by a small, fast LLM that returns a yes/no decision with a reason, so users only see real leads. Token cost is tracked per search. [Claude Haiku, GPT-5 nano, Solr]

eBay inventory and profit automation (HG Garage)
Hourly eBay sync that reconciles real fees, shipping and refunds, and calculates cost of goods for every part automatically, so profit per sale is always correct. [Django, Redis/RQ]

AI document intake and billing for a compliance platform (CE Tax)
Uploads are scanned, read, sorted and filed by AI, with sensitive data redacted and anything uncertain sent to a person. Invoices only go out after an engagement letter is signed. [Django, React, Stripe]

No-code agent builder for researchers (DARE, Carnegie Mellon)
Researchers build agents in a form and connect them on a drag-and-drop workflow canvas. Pinecone or self-hosted Weaviate, depending on where the data must live. 1,000+ researchers. [Django, React Flow, Claude, Gemini]

WHAT I BUILD WITH AI and agents: OpenAI, Claude, Gemini, LangChain, LangGraph, LiteLLM, MCP Voice: Retell AI, Twilio, Whisper, Deepgram RAG: Pinecone, Weaviate, pgvector Stack: Python, Django, FastAPI, React, Next.js, PostgreSQL, Redis, AWS Integrations: Salesforce, HubSpot, Zoho, WhatsApp Business API, Stripe, n8n