Startup Due Dil Turns VC Paperwork Into a 10-Minute Flag Report

Vlad Tislenko, a partner at SMRK VC, has launched Startup Due Dil, a web app that turns a startup pitch deck into a structured due diligence report in roughly 10 minutes. Nine specialist AI agents plus an orchestrator called Oracle pull claims from uploaded materials, cross-check public sources, and mark findings with red, yellow and green flags.

The tool began as an internal helper for Tislenko’s own deal work at the Ukrainian fund. It cost less than €1,000 (about $1,000) to build between May and August 2026. It is already in use at SMRK and under test by other investors after scanning about 100 startups.

That path from personal workflow to shared product is deliberate. Tislenko wanted a first pass that matched how a working partner already read decks, then opened the same pass to peers who face the same volume of early-stage paper.

Ten Agents Split the Work

Users start by uploading a pitch deck. They can add financials, a cap table, legal papers, comments or links for a deeper pass. The system extracts text and visuals, then hands pieces to specialist agents that cover distinct slices of analysis.

• Pitch deck as the core upload
• Financials and traction detail when available
• Cap table and ownership papers
• Legal documents and licensing material
• Comments or links that add context for a deeper pass

Each extra file gives the specialists more to cross-check against public sources. A thin upload still produces a report; a richer pack tends to surface sharper flags.

Oracle can send an agent back with extra context if its first pass looks thin. The finished report carries source citations so an investor can open the underlying pages. Tislenko described the agents as working hard across those ten minutes.

The split matters because early diligence is rarely one question. Team fit, market math, cap table shape and regulatory exposure fail in different ways. Routing each slice to a named agent keeps the report readable when several issues appear at once.

Built for Under a Thousand Dollars

Tislenko used Claude Code, ChatGPT Codex, a DigitalOcean server and Cloudflare protection. The project continues his 2023 Pitch Deck Analyzer, a ChatGPT tool that drew more than 10,000 users who wanted feedback on presentations.

1. 2023 – Pitch Deck Analyzer launches as a ChatGPT tool and passes 10,000 users seeking presentation feedback.
2. May 2026 – Build work on Startup Due Dil begins; an arXiv paper in the same month describes a related multi-agent diligence pattern.
3. May to August 2026 – Full stack assembled for less than €1,000 on Claude Code, ChatGPT Codex, DigitalOcean and Cloudflare.
4. After launch – SMRK runs the tool internally; outside investors join tests after about 100 startup scans.

• $49 for a single report (taxes included)
• $449 per month Angel plan for up to 10 reports
• $999 per month Fund plan for up to 25 reports
• Payments via Paddle (about 8% commission)

SMRK uses the service free. Tislenko’s stated goal is to make the product self-sustaining so profits can fund further experiments. He told Tech.eu he built it partly to stay current with AI and give better advice to portfolio companies such as Esper Bionics, Deus Robotics, Osavul, Ability.ai and Principle.

The same partner background and earlier products sit on the SMRK site that describes a Ukrainian fund specialized in venture capital for IT startups at seed and Round A, typically writing $0.5-1.5 million checks.

Price tiers mirror how those checks arrive. A single $49 report suits a one-off look. The Angel and Fund plans match investors who see steady deal flow and want the same structure on every deck without rebuilding the workflow each time.

Flags That Catch Market Fiction

Key findings land in three colors. Red flags mark material risks, contradictions or claims that public data do not support. Yellow flags mark points that need clarification with the founders. Green flags mark verified positives.

One pattern shows up often. Early-stage founders overestimate market size. “Quite often, Startup Due Dil cannot verify those numbers,” Tislenko said. After years of reading decks he can spot the soft spots quickly. When the tool misses a nuance he updates the relevant agent and reruns the analysis until it catches the issue. He also folds in feedback from the SMRK team and outside investors testing the system.

That feedback loop is the second-order pressure. Founders who know a structured checker will run their materials against public sources have reason to tighten numbers before they send the deck. Investors who receive a flagged report still have to decide which yellows to pursue and which greens to trust, but the first pass no longer consumes hours of manual search.

Color coding also sets a shared language inside a fund. Partners can scan reds first, assign yellows for founder calls, and treat greens as provisional until a human confirms the underlying source. The report does not close debate; it orders it.

Privacy and the Model Providers

Due diligence files contain sensitive financial, legal, ownership and fundraising detail. Tislenko made data security a design priority. Admin dashboards hide private workspace content. Developers do not see user files or reports by default. Users can delete a full run.

Processing still routes through third-party models. The live version uses the OpenAI API; the architecture also supports Anthropic. OpenAI may retain API inputs and outputs for up to 30 days. Tislenko said he designed the system so that OpenAI or Anthropic do not train on customer materials. He also does not train on them himself. He receives logs only for support and updates agents only after direct customer feedback.

Neither do I. I don’t have access to users inputs and outputs, I get logs so that I can help users resolve customer support issues, and I train the system only after direct customer feedback to me.

Vlad Tislenko, Partner at SMRK VC, Tech.eu interview

He still admits the system can make mistakes. Multi-agent design and Oracle’s quality bar reduce but do not eliminate the risk that an AI states something false with confidence.

The 30-day retention window on the provider side is why deletion controls and hidden admin views matter in the product itself. Investors keep the ability to wipe a run even when model vendors hold transient API logs under their own policies.

What Changes for Founders and Funds

Traditional early diligence can stretch from several hours to several days of finding, verifying and structuring information. A 10-minute first cut does not replace that work. It compresses the mechanical layer so partners spend more time on conversations that still matter.

“The goal is not to replace the investor, but to provide a more complete evidence base for decision-making and reduce the risk of overlooking an important signal,” Tislenko said. Reference calls, founder meetings and judgment stay human. The same stands for the founder side: a one-off run before a raise can show readiness gaps that would otherwise surface in the first investor call.

Academic and commercial work is already moving the same direction. A May 2026 arXiv paper lays out a multi-agent framework for VC due diligence that also combines LLMs with live retrieval and structured reports. Other vendors sell tech-only diligence or continuous monitoring. Startup Due Dil’s packaging is the full early-stage stack plus an interactive report a user can keep querying and re-triggering after the first pass.

On X, builders have posted short tutorials for spinning up custom diligence agents with current AI SDKs in roughly the same time window the product claims. The packaged version adds the Oracle loop, the flag system, fund-level customization of thesis and check size, and the privacy claims that a one-off script usually lacks.

Oracle Keeps Weak Passes From Shipping

Oracle is more than a final formatter. It plans the run, watches completeness, and sends thin agent output back with extra context before the report is locked. That loop is how nine specialists stay aligned on one deck instead of drifting into nine partial memos.

When an agent returns a weak pass, Oracle can rerun it rather than leave a hole in the flag set. Citations stay attached so a partner can open the same public pages the system used. The design treats incomplete coverage as a process failure, not a quiet omission.

Tislenko’s own edits follow the same pattern. A missed nuance becomes an agent update, then a fresh run, until the issue shows up in the colors. SMRK feedback and outside tester notes feed that cycle without opening user files to developers by default.

For funds writing $0.5-1.5 million seed and Round A checks, the practical gain is consistency. The same slices get asked every time. Thesis and check-size customization can tilt emphasis, yet the core roster still covers team, market, product, traction, ownership and legal risk in one pass.

Founders Gain a Pre-Raise Readiness Check

The same report that helps investors also works as a mirror for founders. A single $49 run before outreach can reveal which market figures lack public support and which traction claims need cleaner evidence. Gaps that once appeared in the first partner meeting can surface in private first.

Portfolio context shaped that use case. Tislenko built the tool partly to advise companies such as Esper Bionics, Deus Robotics, Osavul, Ability.ai and Principle with sharper prep, not only to speed SMRK’s own screening. A structured checker gives founders a rehearsal against the standards a fund already applies.

Yellow flags become a punch list. Red flags mark claims that need repair or honest narrowing before the deck circulates. Green flags still require human judgment, yet they show where the story already holds up against open sources.

None of that removes the raise itself. Reference calls and live meetings remain the proof layer. The pre-raise run only shortens the distance between a draft data room and a deck that survives the first structured read.

Still the People Work

Tislenko keeps returning to the same limit. Ten minutes is dramatically faster, yet the tool is not meant to stand alone. Investors still need to talk to people who know the founders and walk the flagged points through with the entrepreneur. He expects similar systems will eventually help founders assemble cleaner decks and data rooms. “But ultimately, it will still be a people’s game.”

The product site markets the service as AI Due Diligence in 10 Minutes made by a VC for effective inference. Pricing and the open agent list sit there. Tislenko’s own partner background and earlier products explain why a working investor chose to productize the grind rather than keep it internal forever.

For now the system sits inside SMRK’s process and a small tester group. Its second-order effect is already visible in the feedback cycle: every missed nuance becomes a tighter agent, every unverifiable TAM claim becomes a yellow or red that forces a conversation. That is the practical shift a sub-$1,000 tool can force while the final call stays with the people who still have to look the founder in the eye.

Speed changes the calendar, not the responsibility. Partners still own the decision after the flags are read, the sources are opened, and the founder has answered the yellows in person.