01 / PROJECTRetrieval & RAG
Multi-Tenant Retrieval & RAG Service
Qdrant, FastAPI, tenant-scoped retrieval and semantic caching; retrieval responses rather than unverified generated answers.
Retrieval engineering · tenant isolation
INDEPENDENT ENGINEERING & RESEARCH PORTFOLIO / 2026
AI engineering,
built to be explored.
Production-grade presentation of hands-on AI engineering work: retrieval and RAG, agent orchestration, Transformer architectures, model fine-tuning, distributed training and parallel inference.
Each project has a direct website, source code and README. The portfolio interface is production-grade in design; model-validation and runtime claims remain scoped to each repository.
SYSTEMS / OVERVIEW001 — 004
ENGINEERINGAI ✳ MLEXPLORE THE STACK
RAG
AGENTS
TRAINING
INFERENCE
PYTHONPYTORCHLANGGRAPH,LANGCHAINMCP
09Verified project repositories
Dedicated GitHub Pages + source links
Dedicated GitHub Pages + source links
04Primary engineering areas
Retrieval · Agents · Training · Scale
Retrieval · Agents · Training · Scale
</>Public source & walkthroughs
Explore implementation details on GitHub
Explore implementation details on GitHub
AREAS OF WORK
Across the AI stack.
Different parts of the workflow, from retrieving context to training and running models.
THE PROJECT DIRECTORY
Projects, implementations and code.
Explore all nine project sites directly. Every card provides a website link and reliable source-code and README alternatives.
Website buttons open each dedicated GitHub Pages project. Source and README remain available when a project page has not yet been deployed.
02 / PROJECTAgents & MCP
Multi-Agent Workflows with LangGraph
StateGraph workflows, Pydantic validation, reducers, MessagesState and reusable HR/IT subgraphs.
Graph orchestration · 18 examples
03 / PROJECTModel training
LLM Fine-Tuning & Preference Optimization
Qwen2.5 LoRA tuning with separate SmolLM-based SFT→DPO and ORPO pipelines, using version-isolated training stacks.
Fine-tuning · SFT / DPO / ORPO
04 / PROJECTModel training
Distributed LLM Training with PyTorch DDP
torchrun workers, DistributedSampler, gradient synchronization and Qwen LoRA training workflows with independent model replicas.
Distributed training · torchrun
05 / PROJECTInference & scale
Parallel LLM Batch Inference
Qwen2.5 batch sentiment inference with Python worker queues, per-device model replicas, JSONL validation and ordered outputs.
Batch inference · worker queues
06 / PROJECTModel training
Transformer Encoder from Scratch
Custom multi-head self-attention, positional encoding, pre-LayerNorm blocks and classification on Yelp-style review data.
Encoder architecture · classification
07 / PROJECTModel training
DistilBERT Full vs. Frozen Fine-Tuning
Full-encoder fine-tuning compared with frozen-backbone classification, weighted F1, confusion matrices and saved checkpoints.
NLP fine-tuning · comparison
08 / PROJECTModel training
Transformer Training & Model Evaluation
DistilBERT-based support-text classification, full and frozen-backbone configurations, evaluation and prediction outputs.
Model training · evaluation
09 / PROJECTAgents & MCP
MCP & LangGraph Market Workflow Simulation
LangChain tools, LangGraph routing and STDIO/HTTP MCP servers for simulated stock quotes, synthetic news and non-executing trade actions.
Tool orchestration · simulated operations
No matching projects
Try another keyword or clear your filters.
Nine curated project websites — direct site, repository, and README links on every card.If a project website is unavailable, use its source link and check that repository’s GitHub Pages deployment.
READ THE IMPLEMENTATION
Engineering you can inspect.
Each repository links to a dedicated project website, source implementation and README. The work covers system architecture, practical AI implementation and transparent testing boundaries.
Open GitHub profile01
Code before claimsSource repositories are linked directly.
02
Clear boundariesScope, dependencies and test boundaries are documented.
03
Reproducible engineeringDependencies, setup and usage are documented in project READMEs.