Comprehensive Career Blueprints & Technical Guides

Engineering Roadmaps & Resource Hub

Master the exact technologies and step-by-step pathways required to land high-paying roles in AI, Machine Learning, Data Engineering, and Full-Stack Software Engineering in 2026.

20 Project Ideas Open-Source References

20 Advanced AI Portfolio Projects You Can Build

A practical project list covering AI Agents, RAG, MCP, LLMOps, fine-tuning, GenAI infrastructure, voice AI, computer vision, and production ML systems. Each idea includes an open-source GitHub reference plus guidance on how to turn it into a portfolio project you can explain in interviews.

How to use this list

Do not simply clone a repository and put it on your resume. Pick a real use case, change the data or workflow, add your own architecture decisions, document trade-offs, measure something meaningful, and deploy a working demo. The GitHub projects below are references and building blocks—not finished portfolio submissions.

Agents & Automation → RAG & Evaluation → LLMOps & Infra → Voice & Vision →

1 AI Agents

Multi-agent Ops Crew

What to build

Build a coordinated team of specialized agents that divide work, share state, call tools, and hand tasks to one another.

Make it portfolio-ready

Design an orchestrator, 3–4 specialist agents, shared state, tool calls, retries, and a final synthesis step. Add logs so you can explain how the system decided which agent acted next.

Skills demonstrated

Agent orchestration, state machines, LangGraph, tool calling, reliability.

2 AI Agents

Autonomous Coding Agent

What to build

Create an agent that receives a coding task, inspects a repository, edits files, runs tests, and iterates until it reaches a valid result.

Make it portfolio-ready

Use a sandboxed repository, planning step, code-search tools, patch generation, test execution, and failure recovery. Track every action in a visible execution trace.

Skills demonstrated

Coding agents, repo navigation, tool use, testing, autonomous loops.

3 Automation

Browser-Automation Agent

What to build

Build an agent that can navigate websites, read pages, fill forms, and complete multi-step browser workflows.

Make it portfolio-ready

Give it a concrete workflow such as research collection or repetitive QA. Add guardrails for confirmation before irreversible actions and capture screenshots/logs for debugging.

Skills demonstrated

Browser automation, agents, web interaction, workflow reliability.

4 MCP

Your Own MCP Server

What to build

Expose your own data or application capabilities as Model Context Protocol tools that AI clients can discover and call.

Make it portfolio-ready

Implement a focused server for files, databases, job data, analytics, or an internal API. Include clear tool schemas, validation, error handling, and a sample client.

Skills demonstrated

MCP, tool schemas, integrations, agent interoperability.

5 RAG

Hybrid RAG Over a Niche Corpus

What to build

Build a RAG system that combines semantic vector retrieval with keyword or metadata filtering for a specialized knowledge base.

Make it portfolio-ready

Use a narrow corpus such as legal, finance, healthcare research, technical docs, or product manuals. Compare dense retrieval, keyword retrieval, and a hybrid approach with measurable evaluation.

Skills demonstrated

Embeddings, vector search, hybrid retrieval, reranking, RAG evaluation.

6 RAG

GraphRAG Knowledge System

What to build

Turn unstructured documents into entities, relationships, communities, and graph-aware answers.

Make it portfolio-ready

Create a corpus where relationships matter. Extract entities and edges, build graph summaries, and compare GraphRAG answers with standard chunk-based RAG.

Skills demonstrated

Knowledge graphs, entity extraction, graph retrieval, RAG.

7 Document AI

Document-AI Ingestion Pipeline

What to build

Create a production-style ingestion layer that converts PDFs and complex documents into clean structured content for downstream AI systems.

Make it portfolio-ready

Handle headings, tables, page structure, metadata, chunking, validation, and storage. Add failed-document reporting and reprocessing.

Skills demonstrated

Document parsing, ETL, chunking, structured extraction, RAG ingestion.

8 Evaluation

RAG / Agent Eval Harness

What to build

Build an automated evaluation pipeline that measures whether a RAG or agent system retrieves useful context and generates grounded answers.

Make it portfolio-ready

Create a small golden dataset, run experiments across prompts/retrievers/models, and report metrics in a simple dashboard or experiment table.

Skills demonstrated

RAG evaluation, test datasets, quality metrics, experiment design.

9 LLMOps

LLM Observability Dashboard

What to build

Track prompts, generations, latency, cost, user feedback, traces, and failures for an LLM application.

Make it portfolio-ready

Instrument a real app and build views for slow requests, expensive prompts, error rates, model comparisons, and user feedback. Show how telemetry drives improvements.

Skills demonstrated

Observability, tracing, prompt analytics, LLMOps, production monitoring.

10 AI Safety

Guardrails + Red-Team Testbed

What to build

Create a test environment that checks an LLM application against unsafe outputs, jailbreak attempts, prompt injection, and policy violations.

Make it portfolio-ready

Build attack cases, define expected safe behavior, run automated tests, and report pass/fail outcomes. Include input filtering and output validation.

Skills demonstrated

Guardrails, prompt injection, red teaming, AI safety evaluation.

11 Fine-Tuning

LoRA / QLoRA Fine-Tune

What to build

Fine-tune an open model efficiently for a focused task using parameter-efficient techniques.

Make it portfolio-ready

Prepare a clean dataset, establish a baseline, fine-tune with LoRA/QLoRA, compare outputs before and after training, and document memory/compute requirements.

Skills demonstrated

Fine-tuning, LoRA, QLoRA, datasets, evaluation.

12 Inference

Quantize + Self-Host an LLM

What to build

Deploy an open LLM behind an API and optimize it for practical inference.

Make it portfolio-ready

Compare model size, quantization, throughput, latency, concurrency, and hardware constraints. Expose an OpenAI-compatible endpoint and benchmark it.

Skills demonstrated

vLLM, inference serving, quantization, latency, throughput.

13 GenAI Infrastructure

Multi-Provider LLM Gateway

What to build

Build one API layer that routes requests across multiple LLM providers and models.

Make it portfolio-ready

Add provider fallbacks, retries, model routing, budget controls, rate-limit handling, and request logging. Demonstrate what happens when one provider fails.

Skills demonstrated

LLM gateways, routing, resilience, cost control, provider abstraction.

14 MLOps

Drift & Quality Monitoring Pipeline

What to build

Monitor incoming data and model outputs for changes that can degrade production quality.

Make it portfolio-ready

Build scheduled reports for feature drift, prediction drift, data quality, and model performance. Trigger an alert when a threshold is crossed.

Skills demonstrated

Data drift, model monitoring, quality checks, MLOps.

15 MLOps

Real-Time Feature Store

What to build

Build a reusable feature layer that serves consistent ML features for both training and online inference.

Make it portfolio-ready

Create offline feature definitions, materialize them to an online store, fetch them in a prediction service, and show how point-in-time correctness prevents leakage.

Skills demonstrated

Feature stores, online/offline features, ML infrastructure, data consistency.

16 AI Agents

Text-to-SQL Agent

What to build

Create an agent that turns natural-language questions into SQL, executes safe queries, and explains the result.

Make it portfolio-ready

Use a realistic schema, schema discovery, query validation, read-only controls, error correction, and result summarization. Add tests for ambiguous questions.

Skills demonstrated

Text-to-SQL, agents, database tools, query validation, structured data.

17 Developer Tools

AI PR-Review Bot

What to build

Build a bot that reviews pull requests and produces actionable feedback on code changes.

Make it portfolio-ready

Integrate with a sample repository or GitHub workflow, analyze diffs, categorize issues, suggest improvements, and avoid noisy comments by adding confidence thresholds.

Skills demonstrated

Code review, GitHub automation, LLM developer tools, CI/CD.

18 Voice AI

Real-Time Voice Agent

What to build

Create a conversational voice application with streaming speech input, LLM reasoning, and low-latency speech output.

Make it portfolio-ready

Measure end-to-end latency, add interruption handling, conversation memory, and a real use case such as support, interview practice, or appointment assistance.

Skills demonstrated

Streaming audio, STT, TTS, realtime agents, latency engineering.

19 Computer Vision

Live Detection / Inspection App

What to build

Build a real-time detection application around a camera or video stream.

Make it portfolio-ready

Choose a useful domain such as PPE compliance, defect inspection, traffic analytics, or inventory. Add confidence thresholds, event logging, and an operator dashboard.

Skills demonstrated

YOLO, computer vision, realtime inference, video processing.

20 AI Agents

On-Chain Autonomous Agent

What to build

Build an autonomous agent that can reason over an on-chain or crypto-oriented workflow while keeping actions controlled and auditable.

Make it portfolio-ready

Focus on read-heavy capabilities first: portfolio analysis, governance monitoring, protocol research, or alerting. Add explicit confirmation for any state-changing action.

Skills demonstrated

Autonomous agents, blockchain integrations, tool calling, auditability.

How to choose the right project

If you want AI Engineer roles

Prioritize multi-agent systems, Hybrid RAG, GraphRAG, MCP, evaluation, observability, and model serving.

If you want ML/MLOps roles

Prioritize fine-tuning, vLLM serving, drift monitoring, feature stores, evaluation, and deployment pipelines.

If you want standout demos

Voice agents, browser automation, coding agents, PR review, and live computer-vision applications are highly demonstrable.

Portfolio checklist before you publish

Clear problem statement and target user.
Architecture diagram and technology choices.
README with local setup and demo instructions.
Evaluation, latency, quality, or reliability metrics.
Screenshots, demo video, or deployed application.
Tests, error handling, and production considerations.
What you changed beyond the reference repository.
Three interview-ready technical trade-offs you can explain.

← Back to Resources