PitchPilot = autopilot for institutional equity research

Inspiration

About PitchPilot

Motivation

In traditional equity research, the bottleneck is not accessing information — it is transforming raw data into a clear, defensible investment thesis.

Analysts spend hours reading 10-K filings, extracting key financial and business signals, forming a market view, and then converting that into charts and presentation materials. While many AI tools can summarize documents, they fail at the most critical step: producing a structured, evidence-backed investment thesis grounded in real company data.

We built PitchPilot to solve exactly this problem.

What PitchPilot Does

PitchPilot is an AI Equity Research Copilot that automates the core workflow of institutional stock pitching.

Instead of generating free-form text, it turns raw inputs — including company information, filings, and research notes — into a structured, investment-grade output pipeline:

Filing-grounded business and industry overview
Evidence-backed investment thesis (market view vs. variant view)
Financial analysis with key metrics (revenue, margins, growth)
Automatically generated charts (valuation, trends, scenarios)
One-pager and IC-ready presentation materials
Source traceability for every output

The key innovation is that PitchPilot does not “hallucinate” a story — it grounds its reasoning in real company materials, especially 10-K-style filings, and produces a structured thesis rather than a summary.

How We Built It

PitchPilot is built as a hybrid research pipeline combining data retrieval, structured reasoning, and modular generation.

1. Hybrid Data Layer

User-provided inputs (links, notes, filings)
Retrieved company context (filing-style content, structured facts)

2. Filing-Grounded Context Construction

Extract business model, revenue segments, and key financials
Align qualitative and quantitative signals from filings

3. Schema-Constrained Generation

Instead of free-form output, the LLM generates a structured JSON pitch:
- Summary
- Investment thesis
- Financials
- Chart plan
- IC Q&A

This ensures consistency, modularity, and reusability across components.

4. Automated Chart & Presentation Layer

Chart plans → Python/matplotlib visualizations
Logo-aware theming → clean one-pager layout
Output structured for direct presentation use

Key Innovation

The most important insight behind PitchPilot is:

A stock pitch is not a summary — it is a structured argument.

Most AI tools stop at summarization.
PitchPilot goes further by generating:

A clear thesis
Supporting evidence
Financial implications
A presentation-ready narrative

All grounded in real data.

Challenges We Faced

1. Preventing Hallucination Generating financial narratives without grounding leads to incorrect or misleading outputs.
We addressed this by anchoring generation in filing-style content and enforcing structured outputs.

2. Bridging Qualitative + Quantitative Data Filings contain both narrative and numerical data.
We had to align business descriptions with financial metrics to produce coherent thesis arguments.

3. Structuring Output for Reuse Free-form text is hard to reuse across UI components.
We solved this by enforcing a schema-constrained JSON format, enabling modular rendering across Summary, Pitch, Charts, and IC Q&A.

4. Making Output Presentation-Ready Beyond analysis, we needed outputs that are directly usable.
We built automatic chart generation and layout packaging to bridge research → presentation.

What We Learned

The hardest part of equity research is not data access, but structured reasoning
Grounding AI outputs in real documents is critical for reliability
Structured generation (JSON-first) is significantly more powerful than free-form text
Automating workflows requires thinking in pipelines, not features