These five projects teach end‑to‑end skills employers want—RAG, agents, multimodal pipelines, and productionization—while producing portfolio artifacts that signal job readiness.
- Knowledge Assistant with Agentic RAG
- Build a QA assistant over your notes/PDFs with hybrid retrieval, citations, and a router that decides between local context and web search.
- Ship features: FAISS/Chroma index, source‑level citations, offline evals for faithfulness/latency, and a simple canary/rollback for updates.
- Goal‑Oriented Multi‑Agent Researcher
- Orchestrate agents for search, analysis, and synthesis to produce a referenced brief; add a memory and an evaluator agent for quality control.
- Ship features: task graph (LangGraph/CrewAI), tool APIs, planning and retry logic, and a final export to PDF or Notion with embedded sources.
- Multimodal Study Companion
- Combine OCR/vision with text LLMs: take photos of notes, extract key points, generate practice questions, and provide stepwise hints.
- Ship features: image captioning/OCR, reading‑level control, TTS, and mobile upload; add accessibility options and bilingual output.
- Speech Tutor or ASR‑to‑Feedback App
- Fine‑tune or adapt a speech model for a target language; provide pronunciation scores and corrective feedback with audio examples.
- Ship features: streaming ASR, prosody metrics, progress dashboard, and a privacy‑first data policy with opt‑in consent.
- CI/CD‑Ready AI Microservice
- Wrap an LLM or classifier behind a FastAPI service with tests, evals, and monitoring; deploy to a free tier or container platform.
- Ship features: unit/e2e tests, eval harness with golden sets, observability (latency/cost), drift alerts, and rollback.
What to include in your portfolio
- Repo with README, architecture diagram, and environment files; demo video (~2 minutes); eval results and a model/prompt card detailing risks and mitigations.
- Live demo link or screenshots, plus a short write‑up connecting features to real use cases and SLOs (latency, cost, accuracy).
30‑day build plan
- Week 1: pick one project; scope MVP; set metrics (accuracy/latency/cost); scaffold repo and data pipeline.
- Week 2: implement core logic and two tests per feature; add evals and a minimal UI; start a model/prompt card.
- Week 3: add observability, retries, and error handling; run ablations; record a demo; gather user feedback.
- Week 4: harden security (secret scanning, PII masking), write docs, and publish; iterate based on evals and feedback.
Bottom line: focus on one agentic RAG app, one multimodal pipeline, and one productionized microservice—together they prove you can design, build, and ship reliable AI systems end to end.
Related
How to scope each project for a 4 week student timeline
Which open source models and tools to use for these projects
Project ideas that require no paid cloud resources
Assessment rubric to grade student AI projects
How to add ethics and data privacy into each project