Skip to content
View joel819's full-sized avatar

Block or report joel819

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
joel819/README.md

What I build for clients

Working tools for real business problems. Each one comes with source code, a README with a quickstart and a test suite, and I can adapt it to your data.

catalog-chat-assistant: two shops, each with its own chat grounded in its catalog
A shopping assistant that answers only from a shop's own product catalog: recommendations, comparisons and "which items are vegan under $5?". Says so when something isn't in the catalog, and each shop's data stays separate.
Chatbot · tool calling · ChromaDB · multi-tenant
docs-rag-chatbot: a chat answering a question with a clickable page citation
Answers questions over company documents, with citations to the source page. Click a citation and the PDF opens at that page. A question the documents don't cover gets an honest "I couldn't find that".
RAG · ChromaDB · FastAPI · Groq
invoice-extractor: an invoice flagged for review because the total does not add up
Turns messy PDF invoices into validated structured JSON. Confidence on every field, and arithmetic errors are flagged for review, never silently fixed.
FastAPI · Pydantic · LLM extraction · validation
product-data-extractor: six product pages extracted with per-field sources
Paste a product page URL and get clean structured data, optionally filled into a Google Sheets row. Standard page data first, an LLM only for gaps, and every field shows its source.
Web extraction · schema.org · Google Sheets
web-flow-bot: a real run where two 503 errors were retried and the confirmation verified
Playwright bot that logs in, fills a form, submits it and verifies the confirmation number. Retries with backoff, a screenshot of every attempt and a JSON run log.
Playwright · retries · Docker · pytest
llm-cost-router: benchmark report from a real run
Sends each prompt to the cheapest model that passes a quality check, escalates only on failure and caches repeats. Prices come from config, and the benchmark reports only what a real run measured.
OpenAI-compatible APIs · routing · caching

Engineering projects

How I build the foundations: auth and billing, safety for AI agents, protocols and tool use. These show how I work rather than a product to buy.

fastapi-saas-backend: a playground showing the premium gate, checkout and webhook replay
JWT auth with rotating refresh tokens, subscriptions, verified and idempotent Stripe webhooks, rate limiting and a full test suite.
FastAPI · JWT · Stripe · SQLAlchemy
agent-spend-guard: a dashboard with a daily cap meter and an approval queue
Spend limits and human approval for AI agents that handle funds, with a tamper-evident audit log.
FastAPI · policy as config · audit log
mcp-server-template: an MCP explorer page
An MCP server exposing a real API as a tool for any agent. Streamable HTTP and stdio from one definition.
MCP · GitHub API · stdio
support-agent: answers tagged with the route taken
Support agent combining document search with live order lookups via tool calling. Every answer shows its route and tool calls.
Tool calling · RAG · routing log
sandboxed-agent: the policy drawn as a fence around the agent
Runs support-agent inside an NVIDIA OpenShell sandbox with an allow-list policy for files and network.
OpenShell · policy as code · Docker

How I build

Easy to run
Each README has a short quickstart, and the AI projects have a demo mode or a free-tier setup, so you can see them working in minutes.
Tested offline
Test suites that need no keys, no model download and no network, so they run anywhere.
Honest about failure
Low-confidence flags instead of silent fixes, hard rejects with the numbers, "I couldn't find that" instead of a guess.

Building

SoverGrid: a routing layer for decentralized compute.

Contact

X (@joel_depin)
Email: joeloyewole13@gmail.com
Hire me on Upwork: https://www.upwork.com/services/product/development-it-an-ai-chatbot-that-answers-questions-from-your-documents-with-sources-2106016975697311649

Popular repositories Loading

  1. c.cpp.address.book.project c.cpp.address.book.project Public

    Forked from mickyjm/c.cpp.address.book.project

    C/C++ Address Book Project (Fall 2015)

    C

  2. ai-agents-for-beginners ai-agents-for-beginners Public

    Forked from microsoft/ai-agents-for-beginners

    10 Lessons to Get Started Building AI Agents

    Jupyter Notebook

  3. joel_cpp joel_cpp Public

    C++

  4. quant-stride-master quant-stride-master Public

    Python

  5. after-you after-you Public

  6. intelligent-trading-companion intelligent-trading-companion Public

    C++