applied-ai-lab
A GitHub-ready portfolio lab for Venkat to build small, useful AI products in 3-6 hour weekend sprints.
The goal is to learn modern AI engineering by shipping demoable projects: agents, RAG, structured outputs, evals, MCP-style tools, memory, multimodal workflows, and model routing. Each project should be simple enough to finish quickly, but polished enough to show on a resume.
Portfolio Rules
Each project in this repo should:
- Be demoable in under 2 minutes.
- Teach one or two current AI engineering concepts.
- Include a
README.md, architecture notes, example prompts, and a tiny eval.
- Run locally with minimal setup.
- Solve a recognizable user problem instead of feeling like coursework.
- Follow the shared cost and prompt caching standard.
Structure
applied-ai-lab/
ROADMAP.md
docs/
COST_OPTIMIZATION.md
templates/
PROJECT_CHECKLIST.md
projects/
01-personal-chief-of-staff/
app/
docs/
evals/
README.md
02-deep-research-briefing/
app/
docs/
evals/
README.md
Project Index
| # | Project | Concepts | Status |
|---|
| 01 | Personal Chief of Staff | provider abstraction, structured outputs, tool use, evals | deployed |
| 02 | Deep Research Briefing Agent | planning, citations, critique loops | local MVP |
| 03 | Ask My Documents | RAG, embeddings, retrieval quality | planned |
| 04 | Personal Context MCP Server | tools, resources, reusable context | planned |
| 05 | LLM Eval Arena | eval design, model comparison, scoring | planned |
| 06 | AI Engineering Manager Copilot | issue summarization, decision support | planned |
| 07 | Meeting-to-Decisions Assistant | transcription, extraction, action items | planned |
| 08 | Browser Workflow Agent | browser automation, task state | planned |
| 09 | Multimodal Product Analyst | image understanding, structured critique | planned |
| 10 | AI Data Cleaning Copilot | schema inference, transform suggestions | planned |
| 11 | Voice Journal Memory Coach | speech, memory, reflection loops | planned |
| 12 | Agent Reliability Dashboard | observability, traces, regression testing | planned |
| 13 | Equity Research Lab | source-grounded financial research, provider abstraction, immutable prediction snapshots, retrospective evals | MVP |
| 14 | Shorts Studio | text-to-video prompts, per-platform metadata, publish-readiness lint, scheduling | local MVP |
Weekend Workflow
- Pick one project from
ROADMAP.md.
- Keep the first version intentionally small.
- Add one screenshot or short demo recording after it works.
- Add 3-5 eval cases before calling it complete.
- Write a resume bullet that explains the engineering concept and user value.
- Complete the project shipping checklist, including cache and cost measurements.
Getting Started With Project 01
cd projects/01-personal-chief-of-staff
python3 -m app.server
Then open http://localhost:8000.
The app works in mock mode with no API keys. Add OPENAI_API_KEY or ANTHROPIC_API_KEY to use a real model.
GitHub Profile
Designed for ongoing additions under geeveeyes.
Project 02
cd projects/02-deep-research-briefing
python3 -m app.server
Open http://127.0.0.1:8002 and choose Load example. See the Project 02 README for API key setup and scope.
Project 13
cd projects/13-equity-research-lab
cp .env.example .env.local
npm install
npm run dev
Equity Research Lab runs in demo mode without API keys and is designed to add SEC, analyst, options, persistence, and retrospective-evaluation providers incrementally. See the Project 13 README.
Project 14
cd projects/14-shorts-studio
python3 -m app plan
python3 evals/run_evals.py
Shorts Studio turns episode prompts into Veo prompts, YouTube/TikTok/Instagram metadata, captions and a posting schedule, with no API keys needed. See the Project 14 README.
