Inspiration

Teams today are drowning in data but starving for answers. Dashboards show what happened, but not why or what to do next. We wanted to build something closer to a real analyst, something that doesn’t just visualize data, but thinks through it and delivers decisions.

What it does

Planera is a general-purpose analytics copilot. You give it your data and a question, and it plans the analysis, executes it, identifies root causes, and returns clear insights with recommended next actions. It works across domains making it a universal interface for data-driven decision-making.

How we built it

We designed Planera as an agentic system with a planner-executor-replanner loop using LangGraph. The LLM generates structured analysis steps, which are executed via SQL (DuckDB) or restricted pandas. Results are iteratively refined and finally converted into a clear narrative. The system is exposed via a FastAPI backend and a Streamlit UI.

Challenges we ran into

Constraining LLMs to produce executable, reliable steps instead of vague reasoning
Handling failures and empty outputs with intelligent replanning
Balancing flexibility (any dataset) with safety and structure
Making outputs transparent and trustworthy, not a black box

Accomplishments that we're proud of

Built a working agentic analytics system, not just a chatbot
Achieved end-to-end automation: question → analysis → insight → action
Created a system that is data-agnostic and extensible across domains
Made the entire reasoning process auditable and visible

What we learned

LLMs are powerful planners, but execution must stay deterministic
Transparency builds trust, users need to see how answers are derived
Constraining the system actually improves reliability and usability
- The real value is not analysis, it’s decision-making!