Stock Pitch Simulator

• 

Inspiration

The buy-side recruiting process is gatekept by stock pitches. Every hedge fund, asset manager, and equity research seat tests candidates on the same skill — but how to construct a strong pitch is hidden knowledge. Target-school students learn it through mentors, alumni networks, and paid expert services like Tegus and GLG. First-generation candidates and career switchers don't have access to any of that. They read the same books (Pitch the Perfect Investment) and get the same generic advice, but they never get to see what a strong pitch actually looks like through the eyes of different funds. The result: first-gen candidates pitch in ways that score well for one firm style and fail badly at another, with no feedback loop to course-correct. Stock Pitch Simulator closes that gap.

Roughly 50,000 candidates each year in the US pursue buy-side roles across hedge funds, asset managers, equity research, and private equity. A meaningful fraction are first-generation, career switchers, or otherwise outside existing target-school networks — and they're the candidates most likely to fail the stock pitch gate not because they're less capable, but because they've never seen what a strong pitch looks like through different firm lenses. Each candidate who lands an offer they wouldn't otherwise have gotten captures substantial economic value (buy-side analyst comp starts in the low-to-mid six figures, with compounding career trajectory) and represents real movement on the diversity-in-finance problem that MLT and similar organizations exist to solve.

What it does

Stock Pitch Simulator gives buy-side interview candidates rigorous, firm-specific feedback on their stock pitches. Users:

Select a target firm style — Long-Only Quality (Pershing, Sequoia), Classic Value (Baupost, Greenlight), L/S Pod (Citadel, Millennium, Point72), Tiger Cub (Coatue, Tiger Global), Event-Driven (Third Point, Elliott), Distressed/Restructuring (Oaktree, Centerbridge), or Activist (Starboard) Write three structured inputs: their Investment Thesis, Catalysts, and Key Risks Submit for evaluation by an AI Portfolio Manager calibrated to that specific firm's investing philosophy Receive a score (0-100), specific strengths and weaknesses, and actionable advice

For users who don't know what good looks like, a guided walkthrough of a strong Costco pitch shows each section with annotations explaining what makes the pitch work. Pre-loaded demos (Professional Apple, Amateur Tesla, Costco Walkthrough) let users see contrast in seconds. The core insight: the same pitch is evaluated entirely differently across firm styles. A growth-narrative thesis that scores 88 for an L/S pod scores 42 from a value PM — mirroring real buy-side recruiting outcomes that first-gen candidates currently navigate blind.

How we built it

Built in Google AI Studio Build mode using Gemini 3 Pro Preview. The technical architecture is intentionally simple — what matters is the system prompt design. The "AI Portfolio Manager" is implemented as a long, calibrated system prompt that maps Sonkin and Johnson's Pitch the Perfect Investment framework into seven firm-specific evaluation rubrics. Each firm style has its own weighting matrix that prioritizes different aspects of a pitch — variant view specificity for L/S pods, margin of safety and asset coverage for value, TAM defensibility for Tiger cubs, capital structure for distressed, operational change levers for activists. The same pitch flows through different evaluation logic and produces meaningfully different feedback. The Costco walkthrough page uses pre-filled example content with annotation callouts on each section explaining what specifically makes that part of the pitch strong — turning a static example into an active lesson.

Challenges we ran into

Scope discipline. The initial plan had idea generation mode, pitch construction mode, adversarial Q&A simulator, voice mode, firm matching with 13F data, primary research scaffolding, news ramp, and peer practice matching. Cutting hard to one focused product was harder than building it.

API quota and billing. Free-tier Gemini limits surfaced quickly, and the "Pay per request" toggle in Build mode doesn't propagate paid status without a properly linked billing account on the API key's parent project. Diagnosing that took longer than it should have.

UX iteration on the Learn mode. Early versions of a chat-based educational flow produced walls of dumped text rather than paced lessons. Two full rebuilds and a pivot to a simpler architecture were needed before settling on the structured input + structured feedback design.

Prompt engineering across firm styles. Getting the AI to genuinely differentiate between firm styles (and not just rewording the same evaluation) required multiple iterations of the rubric. The final version explicitly encodes firm-specific praise patterns and failure modes.

Accomplishments that we're proud of

The firm style differentiation actually works. The same Lamb Weston pitch, run through Citadel-style L/S pod, Baupost-style value, and Pershing-style long-only quality, produces three meaningfully different evaluations, each catching the failure modes that the particular firm style would actually flag. That's the bulk of the value proposition.

The Costco walkthrough is genuinely educational. Annotations don't just describe what was written — they explain why it works as a pitch, in plain language.

Built end-to-end in one weekend by a first-time AI builder, in a vibe-coding environment, without prior frontend experience.

MLT mission alignment is clean. This is a translation layer between investing epistemologies, built specifically to close the access gap for candidates without buy-side networks.

What we learned

The system prompts are the moat, not the code. A well-crafted system prompt with the right domain knowledge embedded does 80% of the work; the wrapper is plumbing.

Vibe coding works, but only with ruthless scope discipline. The initial brainstorm had eight features. The shipped product has four. The cut features make v2 stronger; trying to ship all eight would have left nothing usable.

Domain expertise compounds with AI tooling. The ability to write firm-style profiles grounded in actual co-investment work was the real differentiator, and the AI capability was just the method.

Edge cases reveal product quality. Stress-testing the same pitch across all seven firm styles, including ones the pitch obviously didn't fit (Distressed on an investment-grade name), exposed where the calibration was forcing rubrics versus where it was gracefully acknowledging mismatch.

What's next for Stock Pitch Simulator

Idea generation mode for users who want to break into buy-side but don't yet have a stock pick. Walks users through circle-of-competence filtering, narrative scanning, and latent-thesis surfacing to land on a name worth developing.

Adversarial Q&A simulator. After the pitch is built, the AI plays a hostile interviewer and fires the standard buy-side kill shots — variant view, bear case, exit triggers, edge, sizing, why-hasn't-the-market-figured-this-out.

Voice mode via Gemini Live. Real interviews are spoken, not typed — the conversational pressure matters.

Value-added research scaffolding. Helps first-gen candidates do primary research without target-school networks: geographic channel checks, LinkedIn alumni search assistance, free-tier expert network alternatives, structured question design.

Firm-specific calibration within each style. Style buckets capture 80% of what matters, but specific firms have idiosyncrasies — Pershing's activist lens, Coatue's private/public crossover, Baupost's special-situations comfort. The full version adds these as overlays.

Peer pitch practice matching. Once the AI loop has user retention, matching becomes a feature on top of an existing product rather than a marketplace cold-start problem.

Sustainability path. Stock Pitch Simulator has multiple plausible revenue paths. The most natural Year 1 wedge is institutional licensing to MBA career services offices and finance societies — universities pay an annual fee to provide access to enrolled students preparing for buy-side recruiting, similar to how schools already license CFA and investment banking prep platforms. A second path is direct-to-consumer freemium ($19-49/month for unlimited evaluations, voice mode, and adversarial Q&A), modeled on existing finance interview prep tools. A third is partnership with diversity-in-finance organizations (MLT, SEO, Toigo, Out Investors) that subsidize access for their fellows as part of broader career programming. The product cost structure is dominated by Gemini API spend, which scales linearly with usage and is well covered even by modest paid conversion rates.

Built With

• claude
• gemini