Skip to content
View bartuonder's full-sized avatar
๐Ÿ’ญ
Bartu ร–nder
๐Ÿ’ญ
Bartu ร–nder

Organizations

@AIOSECENG

Block or report bartuonder

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 donโ€™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 userโ€™s behavior. Learn more about reporting abuse.

Report abuse
bartuonder/README.md

๐Ÿ‘‹ About me

I'm a 3rd-year Software Engineering student (full scholarship, GPA 3.78/4.00) who is fascinated by every corner of AI. My long-term goal is to become an AI researcher / scientist: to do a PhD in AI in the USA and help shape where the field goes next.

I care about both halves of the job:

  • ๐Ÿ”ฌ The science: ML/DL theory, the math underneath, reading papers and re-implementing them until I really understand them.
  • ๐Ÿ› ๏ธ The engineering: training models, serving them, tracking and deploying them, and turning them into real products.

"If I can't build it from scratch, I don't understand it yet." ๐Ÿงฉ

class BartuOnder:
    role      = "AI/ML Engineer  โ†’  AI Researcher"
    focus     = ["Deep Learning (PyTorch)", "LLMs & Transformers", "RAG & Agentic AI", "MLOps"]
    building  = "end-to-end, scalable, AI-powered SaaS products"
    next_step = "PhD in AI ๐Ÿ‡บ๐Ÿ‡ธ"

    def philosophy(self):
        return "Read the paper. Implement it. Break it. Ship it."

โญ Flagship Project: AxiomForge

Replicate machine-learning claims with statistics, not with a language model. ๐Ÿ”ฌ

An autonomous AI research agent that reads a claim (or an arXiv link), tests it on many real datasets inside a sandbox, and lets Python, not the LLM, decide the verdict. The research agent then proposes its own follow-up hypotheses and tests them on datasets it has never seen.

AxiomForge demo: summary, evidence ledger, research agent session and live progress
70-second walkthrough recorded from the real console.

๐Ÿ’ก The idea: LLM agents write fluent conclusions but are unreliable judges of evidence. AxiomForge splits the work: the model plans experiments and writes prose, while a deterministic statistics layer computes every verdict and ties every number to an evidence ledger.

๐Ÿงช Question ๐Ÿ“Š Measured result
Does it get known claims right? 10 / 10 established comparisons judged as expected, on 9 real datasets
Does it invent differences? Decisive verdict in only 2.1 % of simulated worlds with no true effect; 0 / 45 real-model comparisons significant when the only difference was the random seed
Does the research agent inflate discoveries? Family-wise error 3 %, versus 24 % for an agent that selects and tests on the same data and 89 % for uncorrected testing
A real research session 6 models ร— 16 datasets: 16 hypotheses registered โ†’ 15 confirmed, 0 reversed, 1 not confirmed. The "winner's curse" overestimated the lead of tree ensembles ~3ร— on the exploration data, so only confirmation datasets are reported

๐Ÿ”‘ What makes it interesting

  • ๐Ÿ”’ Sealed confirmation data: the hypothesis generator never sees the confirmation datasets. Hypotheses are frozen with a SHA-256 digest before testing, and python -m app.research verify detects tampering.
  • ๐Ÿ“ Real statistics: Wilcoxon signed-rank, Friedman, Holm and Benjamini-Hochberg corrections, with the dataset (not the split) as the unit of replication.
  • ๐Ÿšง LLM output is policed: the paper text is composed from ledger values and checked by 17 deterministic rules that reject overclaims.
  • ๐Ÿ“ Honest limits: conservative, low power under 1 ROC-AUC point, and the real session reproduces a known finding (Grinsztajn et al., NeurIPS 2022) rather than claiming a new one.

Repository Technical report Demo


๐Ÿงญ What I work on

Click a section to expand ๐Ÿ‘‡

๐Ÿ”ฌ ย Research & Theory

Reading papers, deriving things by hand, and re-implementing architectures from scratch to see how they really work.

See ๐Ÿงช Research & From-Scratch Implementations below.

๐Ÿง  ย AI / ML / GenAI
Area Topics & tools
Deep Learning PyTorch, CNNs, Vision Transformers (ViT), MobileNet, DeBERTa, transfer learning, mixed-precision training
NLP & LLMs Hugging Face Transformers, fine-tuning, semantic search & embeddings, structured outputs
Computer Vision OpenCV, image classification, object detection (YOLO), face recognition & emotion analysis
GenAI / RAG RAG pipelines, vector databases (Chroma, Qdrant), chunking & retrieval evaluation (Hit@k, MRR)
Agentic AI Tool calling, ReAct loops, planning, memory, reflection, LangChain / LangGraph (human-in-the-loop interrupts, checkpoints), autonomous research agents, LLM output guardrails
Evaluation & statistics Wilcoxon, Friedman, Holm / Benjamini-Hochberg corrections, false-positive calibration, power simulations, experiment reproducibility
Data science NumPy, SciPy, Pandas, Matplotlib, Seaborn, scikit-learn, XGBoost / LightGBM / CatBoost

AI stack

๐Ÿš€ ย MLOps & Deployment

Models that only live in notebooks don't count. I train โ†’ track โ†’ containerize โ†’ serve โ†’ monitor.

  • ๐Ÿ“ˆ Experiment tracking: MLflow
  • ๐Ÿณ Containers: Docker, Docker Compose
  • โ˜๏ธ Cloud: AWS
  • โšก Inference: async inference queues (Celery + RabbitMQ), ONNX Runtime, Redis caching
  • ๐Ÿงฐ Sandboxed execution: running agent-generated experiment code in isolated sandboxes (E2B)
  • โœ… CI & quality: GitHub Actions, Pytest, mutation checks, Playwright browser tests
  • ๐Ÿ—„๏ธ Model & artifact hosting: Hugging Face Hub

MLOps stack MLflow

๐Ÿ—๏ธ ย Backend & SaaS Development

Production-minded backends that wrap AI models in clean, tested, scalable services.

Layer Stack
API FastAPI, Starlette, Pydantic
Async & queues Celery, RabbitMQ, Redis
Data PostgreSQL (+ pgvector), SQLite, SQLAlchemy, Alembic, Redis, Vector DBs
Frontend Next.js, Tailwind CSS (dashboards for my SaaS products)
Quality Pytest, Clean Architecture, Dependency Injection

Backend stack

๐Ÿ›ก๏ธ ย Security

I build AI products, so I also care about securing them.

  • ๐Ÿ” Web & API security: OWASP Top 10, OWASP Top 10 for LLM Applications, API security
  • ๐Ÿ”‘ AuthN / AuthZ: OAuth, JWT, API keys
  • ๐Ÿšฆ Hardening: rate limiting, reverse proxy, basic pentesting
  • ๐Ÿงฑ AI security: prompt-injection defense, PII redaction (Sentinel-Shield)
๐Ÿ’ป ย Languages & Tools

Languages and tools


๐Ÿงช Research & From-Scratch Implementations

Learning by building: no black boxes, paper-faithful code, and tests where it matters.

Paper-to-code implementation of the original 2017 Transformer, written to match the paper (Post-LN, sinusoidal PE, tied embeddings, label smoothing, Adam warmup, beam search with length penalty) rather than wrapping nn.Transformer. Includes tests and a paperโ†’code architecture map.

๐Ÿค– GPT-2 (124M)

Clean from-scratch GPT-2 small: masked multi-head self-attention, FlashAttention via SDPA, loading of official pretrained weights, bf16 mixed-precision fine-tuning with cosine LR, grad accumulation and clipping, and temperature / top-k sampling.

๐Ÿ•ต๏ธ AI Agents From Scratch

Agents in pure Python, with no frameworks: LLM call โ†’ tool calling โ†’ ReAct agent loop โ†’ memory โ†’ plan-and-execute โ†’ reflection, plus evals and tests. The goal: show there's no magic in an agent, just a loop.

Classic estimators in NumPy only: linear/logistic/softmax regression, decision trees, random forests, second-order gradient boosting, dual SVM (SMO), GMM, GDA, naive Bayes, k-means, PCA and a one-hidden-layer MLP. Every model is derived from an explicit objective.

๐Ÿ•ธ๏ธ Neural Networks From Scratch

A small neural-network library with layers, activations, losses and manual backprop, validated with gradient checking and demos on XOR, regression and classification.

๐ŸŽญ ShakespeareGPT

A decoder-only GPT trained on Shakespeare's poetry to emulate the Bard's linguistic style and meter: dataset, model, training and generation scripts.

Also: ๐Ÿ“š Paper-to-Code (a growing home for AI/ML/DL paper implementations) ยท ๐Ÿ”ฅ PyTorch-General-Codes (CNNs, ViT, transfer learning, TensorBoard, YOLO) ยท ๐Ÿ“Š Kaggle notebooks


๐Ÿš€ Featured Projects

End-to-end AI products, from model to deployment.

Project What it does Stack
๐Ÿ“‘ Veridex Extractive contract risk QA SaaS. Fine-tuned DeBERTa-v3 on CUAD โ†’ 87.96 F1, tracked with MLflow. Inference isolated from the API via Celery/RabbitMQ, Redis caching, JWT/API-key auth, Next.js risk-card dashboard. ๐Ÿค— Model PyTorch DeBERTa-v3 FastAPI Celery Redis PostgreSQL Next.js MLflow Docker
๐Ÿ›ก๏ธ Sentinel-Shield AI security firewall for LLM apps: PII redaction and prompt-injection blocking. Deployed on AWS, with JWT auth, Redis rate-limiting and a user dashboard with API keys and stats. FastAPI Starlette SQLAlchemy AWS Redis Next.js
๐Ÿฅ— MacroFit AI Estimates calories and macros from food photos with a Vision Transformer, then builds nutrition tracking and workout plans. AI layer isolated via dependency injection. PyTorch ViT FastAPI Next.js PostgreSQL Docker
๐Ÿ–ผ๏ธ Vision Classifier API ViT image-classification platform with an async task queue so heavy inference never blocks the server. JWT-secured and horizontally scalable via Docker. PyTorch Hugging Face Celery RabbitMQ JWT Docker
๐ŸŒฟ LeafLens MobileNetV3 detecting 38 plant diseases at ~97% accuracy, served as a REST API with request logging. PyTorch FastAPI SQLAlchemy Tailwind
๐Ÿ‘๏ธ DeepGuard AI access-control / mini-SOC: face recognition, age estimation and emotion-based anomaly detection (anger/fear) with real-time event logging. PyTorch Transformers FastAPI
๐ŸŽฎ NextQuest LLM taste-based recommender for games, movies, books, series and anime, with Pydantic structured outputs, top-up rounds for incomplete answers and per-query taste snapshots. LangChain OpenAI Pydantic PostgreSQL Docker
๐Ÿง  DocuMind API Production-grade RAG platform to chat with your documents: async ingestion, vector search, ONNX inference and MLOps tooling. FastAPI PyTorch ONNX Qdrant Celery MLflow
โ˜ฃ๏ธ ToxGate Async Turkish toxicity detection PoC: DistilBERT on ONNX Runtime behind FastAPI, Celery workers and a Redis result cache. DistilBERT ONNX FastAPI RabbitMQ Redis

๐Ÿ’ผ Experience

๐Ÿข Software Engineering Intern, GullsEye Logistics Technologies Inc. (20 days)

Designed and built, end to end, a fully offline AI-powered semantic search engine over enterprise documents.

  • โšก Cut search latency to ~5 ms by optimizing vector matrices with NumPy
  • ๐ŸŽฏ Reached Hit@3 = 90.6 % and MRR = 0.735 using custom chunking and an automated evaluation harness
  • ๐Ÿค— Used Hugging Face MiniLM-L12 embeddings, with a CLI on top of the core engine
  • โœ… Secured the whole project with 164 unit and integration tests (Pytest)

๐Ÿ“Š GitHub Stats


๐ŸŽฏ Currently

  • ๐Ÿ”ญ Re-implementing more landmark papers in Paper-to-Code
  • ๐ŸŒฑ Going deeper into LLM fine-tuning, RAG evaluation and agentic systems (LangGraph)
  • ๐ŸŽ“ Preparing for graduate studies / PhD in AI in the USA
  • ๐Ÿค Open to research collaborations, internships and interesting AI projects

๐Ÿ“ซ Let's connect! ย  LinkedIn ยท Hugging Face ยท Kaggle ยท Email

footer

Pinned Loading

  1. AxiomForge AxiomForge Public

    AxiomForge - Autonomous AI Research Agent

    Python 1

  2. Mabeyn Mabeyn Public

    Multi-tenant support agent you embed with one script tag: cited answers from your documents, tool use with human approval, customer memory and an LLM firewall. FastAPI, LangGraph, PostgreSQL RLS, Nโ€ฆ

    Python 1

  3. Veridex Veridex Public

    AI contract analysis project

    Python 1

  4. Paper-to-Code Paper-to-Code Public

    Index of research papers I re-implemented from scratch in PyTorch: paper โ†’ code maps, verification tests, and reproduced results.

    1

  5. credit-default-risk credit-default-risk Public

    Calibrated credit card default prediction on the UCI dataset: repeated-CV model comparison, cost-based decision threshold, SHAP explanations, and a FastAPI service with a web interface.

    Python 1

  6. machine-learning machine-learning Public

    Machine Learning

    Python 1