Good evening

Oussama Gader: Full stack & product engineer based in Monastir, Tunisia.

I ship AI agents, backend services, APIs, and web UIs end to end, as a startup co-founder and independent builder. Have a look at my work, the projects I build on my own, or the stack behind them.

919 contributions since 2024
Monastir, Tunisia (Remote)
Local time:
oussamagader2023@gmail.comgithub.com/gaderOussama

Work

(3)

Where I've been building.

Sidi Alouane News

Founder · Grew it from 0 to 280K+ followers
2018 – now

Projects

(5)

Things I build on my own.

Bud Bud live

An accountability AI companion on Telegram — it remembers what you said you'd do, holds onto your ideas, and learns your patterns well enough to flag it when something's off, built around natural Tunisian Darija.

  • AI
  • Telegram
  • Gemini
  • Conversational AI

Engineering depth

  • Repositioned from a general AI companion — which nobody came back to — into an accountability companion: remembers your tasks, reminds you what you said you'd do, stores your ideas, and learns your patterns to flag when something's off.
  • Chose Telegram over WhatsApp: WhatsApp's own bot ecosystem would make the product dependent on a platform that could cut bot access at any time.
  • Chose Gemini over Gemma for the Darija conversational core — multimodal, with a paid tier to run it sustainably.
  • Whisper's zero-shot error rate on Tunisian Arabic is very high, so Tunisian-specific speech models were compared before committing to fine-tuning.

Tempo hackathon

Multimodal video shot search: video editors describe a shot by what's on screen or what's said, and get ranked, timecoded results. Fuses SigLIP 2 visual embeddings, Whisper transcription, Florence-2 captions, OCR and entity extraction into one hybrid ranking, with a prototype built toward an Adobe After Effects panel.

🏆 Palete AI Creative Experience Award — Come Build with AI Hackathon (GOMYCODE × NVIDIA), Sept 2026

  • AI
  • Python
  • PyTorch
  • Hugging Face
  • SigLIP 2
  • Whisper
  • Florence-2

Engineering depth

  • Pipeline fuses SigLIP 2 for visual matching, Whisper for speech, Florence-2 for scene understanding and OCR for on-screen text into one hybrid ranking step.
  • Built with one teammate, a professional video editor who defined the real workflow.
  • Won the Palete AI Creative Experience Award at the GOMYCODE × NVIDIA hackathon — with no local GPU.

Akarwcar live

A live marketplace for property and vehicle listings from one codebase, with a three-role approval workflow.

  • Next.js
  • TypeScript
  • Supabase
  • PostgreSQL
  • Tailwind CSS

Engineering depth

  • Leading the tech as freelance tech lead since May 2026, after it began as a Kweektech client engagement and staying on after leaving.
  • Core design is a prop-driven component system, so one set of components serves both property and vehicle listings.
  • Listings move through a three-role approval workflow: admin, seller, client.

MicroManus live

Deep-research AI agent. Tool calling across five LLM providers, web search, a Stripe credit paywall and a cost dashboard.

  • Next.js
  • Supabase
  • Stripe
  • Vercel
  • Brave Search API
  • Agents

Engineering depth

  • Users bring their own API key for OpenAI, Anthropic, Gemini, Groq or xAI — encrypted with AES-256 at rest and sent only to the provider they picked.
  • A tool-calling agent loop searches the web, fetches pages, and generates PDF reports stored in object storage and linked back in the chat.
  • A per-chat dashboard breaks down cost and token usage by model and by chat.
  • Stripe paywall grants credits through a verified webhook.
  • Built and shipped solo. Open source, MIT.

Lantern in progress

RAG-powered reading companion that talks about books in a literary character's voice, streamed over SSE.

  • RAG
  • pgvector
  • SSE

Engineering depth

  • Readers often feel alone after finishing a book, or miss a point the author meant to land — Lantern lets you talk to the book itself, going deep on its characters and world.
  • RAG-powered: retrieves from the book's own text so the voice and facts stay grounded in it.
  • Responds in a literary character's voice, streamed over SSE.

Open source

(6)

Public repositories, pulled live from GitHub.

Writing

(1)

Notes on building products and what I learn shipping them.

Stack

(43)

Tools I use every day. Means, not identity.

01 Build

  • Next.js
  • React
  • TypeScript
  • Node.js
  • Python
  • FastAPI
  • Tailwind CSS

02 Ship

  • PostgreSQL
  • Supabase
  • Redis
  • Convex
  • Stripe
  • Twilio
  • Docker
  • Git
  • Vercel

03 Monitoring

  • Grafana

04 AI & data

  • Data Science
  • RAG
  • Agent Systems
  • Tool calling
  • Multi-provider LLMs

05 Machine learning

  • Linear & Logistic Regression
  • Decision Trees
  • Random Forest
  • XGBoost
  • SVM
  • k-NN
  • K-Means
  • PCA
  • Neural Networks
  • CNN
  • RNN / LSTM
  • Transformers

06 AI tooling

  • Claude Code
  • Codex
  • Trae
  • Kimi
  • GLM

07 Think & brainstorm

  • Notion
  • Obsidian
  • a notebook & a pen
  • my brain, and Claude's

Education

(2)

ISIMM, Université de Monastir.

M.Sc. Data Science

ISIMM · Monastir
2025 – now

Licence, Computer Science

ISIMM · Monastir
2022 – 2025

Have something interesting in mind?

I'm always open to new ideas, collaborations, or just a good conversation.

Get in touch