EquityLens

EquityLens: Your autonomous AI associate. It reads pitch decks, fact-checks market claims, and spots red flags in seconds. Stop drowning in noise; start spotting the next unicorn.

Comment

Inspiration

The idea for EquityLens was born from two conflicting chapters of my life.

First, during my VC Fellowship at Metvy, I saw the "analyst bottleneck" firsthand. I spent countless nights manually writing investment memos, cross-referencing market sizes, and digging through 50-page PDFs just to find one red flag. The due diligence process was slow, manual, and prone to human error.

Second, as a 2x SaaS Founder (building Guildup.club and whomr.in), I experienced the other side: the frustration of waiting weeks for a VC reply, only to get a generic rejection because they didn't have time to properly understand my deck.

I realized that the bottleneck wasn't a lack of capital—it was a lack of attention bandwidth. I built EquityLens to solve this by giving every investor an autonomous "AI Associate" that works at the speed of software.

What it does

EquityLens is not a chatbot; it is a multi-agent Due Diligence System. It acts as an autonomous VC Associate that performs deep analysis on startup pitch decks in seconds.

Users upload a Pitch Deck (PDF), and our system triggers three specialized AI Agents:

  1. Agent A (The Market Detective): Reads the startup's market claims and autonomously searches Google to verify them against real-time data (e.g., "Is the EdTech CAGR actually 15%?").
  2. Agent B (The Risk Auditor): Uses Gemini 1.5 Pro's massive context window to deep-dive into the legal/financial text, spotting "AI Wrapper" risks, vague financials, or cap table issues.
  3. Agent C (The General Partner): Synthesizes these conflicting reports into a final Investment Memo with a calculated "Deal Score" (0-100) and a hard "Invest/Pass" verdict.

How we built it

We leveraged the cutting-edge Google AI Stack to build a true Agentic workflow:

  • The Brains: We used Gemini 2.0 Flash for high-speed text extraction and Gemini 1.5 Pro for deep reasoning, leveraging its 2M+ token window to ingest entire data rooms.
  • The Grounding: We implemented Google Search Grounding via the GenAI SDK. This allows our agents to "fact-check" hallucinations, ensuring that market data comes from live web sources, not just training data.
  • The Backend: Built on Node.js (my core expertise) and hosted on Google Cloud Run. We used parallel processing (Promise.all) to run the Market and Risk agents simultaneously to reduce latency.
  • The Dev Environment: The entire application was coded inside Project IDX, which streamlined our integration with the Gemini API and Firebase Data Connect.

Challenges we ran into

The biggest technical hurdle was "The Sycophant Problem." Early versions of the AI were too nice—they would just summarize the pitch deck and praise the startup.

  • Solution: We had to engineer "Cynical System Prompts." We explicitly instructed Agent B to act as a "ruthless auditor" and trained it with few-shot examples of bad deals.
  • Hallucinations: Initially, the AI would invent market numbers. We solved this by strictly binding Agent A to Google Search Grounding, forcing it to cite sources for every number it produced.

Accomplishments that we're proud of

  • True Multimodality: We aren't just parsing text; we are feeding raw PDF visuals to Gemini. The model can interpret graphs and charts in the pitch deck that text-only parsers would miss.
  • Real-Time Fact Checking: Successfully implementing the "Search Tool" so the AI knows when a founder is inflating their TAM (Total Addressable Market).
  • Speed: Reducing a process that usually takes a human analyst 4 hours down to 35 seconds.

What we learned

  • Agents > Chatbots: We learned that the future of AI isn't chatting with a bot; it's assigning a bot a job and letting it go work.
  • Context is King: My experience writing memos at Metvy taught me that context matters. Gemini's long context window allowed us to feed it not just the deck, but also founder bios and competitor data without hitting token limits.

What's next for EquityLens

We plan to double down on the "Agentic" capabilities:

  1. LinkedIn Cross-Reference: Automatically checking founder backgrounds against their real LinkedIn profiles to verify "ex-Google" or "ex-Meta" claims.
  2. Term Sheet Generator: Allowing the "General Partner" agent to draft a non-binding term sheet if the Deal Score is above 90.
  3. Crunchbase Integration: Connecting to live private market databases for even deeper competitive analysis.

Built With

Share this project:

Updates