Erudio Bio, Inc.

posted: 27-May-2026 & updated: 28-May-2026

Erudio Bio Seed Pitch Deck — Consolidated Panel Review

Deck: Erudio Investor Pitch 2026 05 25.pdf (12 slides, 4.1 MB, dated 2026-05-25)
Panel: Narrative reviewer • Numbers reviewer • Devil’s Advocate
Date of review: 2026-05-27
Drafts: /tmp/pitch-review/narrative.md, /tmp/pitch-review/numbers.md, /tmp/pitch-review/devils.md

Process note. Each reviewer received the 12 rendered slide PNGs plus the text extract, and was told the others would push back via SendMessage. In execution the sub-agent sandbox blocked both file writes and inter-agent messaging, so each reviewer produced their draft independently and the orchestrator (this synthesis) reconciled conflicts after the fact, using each reviewer’s explicit hand-off notes to the other panelists. Where the three reviews disagreed, the disagreements are surfaced in §3 rather than blended away.

1. Headline Verdict

Consensus: not investable as written, but the underlying science and team are credible enough that a revised deck could be. The same five-or-six gaps surface from every reviewer’s seat: no ask, no traction, no problem statement, no GTM, statistically thin hero data, three businesses pitched as one.

Reviewer Grade Verdict
Narrative C- Two strong assets (bioTCAD/EDA frame, 8/8 vs 1/8 chart) buried under 5 architecture slides; three-market sprawl with no GTM, no ask.
Numbers D+ Tech vision wrapped around a single n=8 anecdote whose 95% CI overlaps with the comparator’s; no sourced TAM, no ask, no regulatory plan.
Devil’s Advocate REVISIT IN 6 MONTHS Three companies in a seed-shaped trench coat; “TCAD-for-pharma” promise invokes Schrödinger but never names them; no ask, no traction, no commercial hire.

If you sent this deck to a triage-driven seed partner tomorrow, all three reviewers expect a pass — not because the underlying technology is wrong, but because the deck does not contain enough information to write a memo.

2. Where All Three Reviewers Agreed (the unanimous fixes)

These are the items every reviewer flagged from their own seat. Fix these first; they cost nothing to disagree about.

  1. No ask, no use of funds, no milestones. Round size, pre-money, runway, 12-/18-/24-month milestones, prior funding — none present. Numbers calls this “fatal.” Devils calls it “I cannot write a memo I cannot price.” Narrative calls it “the deck just stops.”
  2. No problem statement / no quantified pain. Slide 2 jumps straight to a stack diagram; the reader never feels urgency. Insert a problem slide before any capability slide.
  3. No competitive landscape, no incumbents named. Schrödinger, Recursion, Insilico, Iambic, Isomorphic Labs, Chai — zero appear. The Slide 6 “bioTCAD = EDA” frame invites the Schrödinger comparison and then ducks it. A competitive slide is non-negotiable.
  4. Three markets, one team, one round → pick one wedge. Drug development, biodefense, and clinical diagnostics each have different buyers, sales cycles, regulatory paths, and unit economics. All three reviewers independently said: drop two, focus on the third for this raise.
  5. The Synopsys / Samsung / SK Hynix logos on Slide 8 read as partnerships but are almost certainly founder employers. All three flagged this. Either label them (“founder pedigree”) or remove them — leaving them ambiguous is worse than either fix.
  6. “Initial interests from SNUH and Keimyung” is not traction. All three want it quantified: LOI? signed pilot? paid POC? dollar amount? date?
  7. “Error-free multiplexing” (Slide 3) contradicts “error removal” (Slide 5). Numbers and Narrative both flagged this. The defensible claim is error suppression / filtering — Slide 3’s absolute language should be softened.
  8. The bioTCAD/EDA framing on Slide 6 is the strongest single idea in the deck. All three reviewers agree. They differ on whether it’s currently earned by the evidence (see §3.1), but they agree it belongs at the front, not buried at slide 6.

3. Cross-Reviewer Conflicts and Synthesis

The reviewers disagreed on four substantive points. Each is resolved below with the orchestrator’s reconciled recommendation.

3.1 Should Slide 7 (“100% vs 12.5%”) be promoted to the cover slide?

  • Narrative says YES. Slide 7 is “the deck’s killer slide, buried at #7. ‘8/8 right where state-of-the-art AI got 1/8’ is the entire pitch in one image. Move it forward.”
  • Numbers says NO. With n=8 (one of which is the anchor used to fit the force field parameters), Wilson 95% CIs are [63%, 100%] for bioTCAD and [2%, 47%] for the legacy tool — the intervals overlap. “100% vs 12.5%” is statistically indistinguishable from “75% vs 25%” at this n. Promoting the chart amplifies the overclaim.
  • Devils piles on. All eight peptides derive from one 18-mer cyclic antiviral parent. The “legacy in-silico” tool is unnamed. If Erudio ran it themselves in default mode, the 12.5% number is unreliable too.

Reconciled recommendation. Promote the bioTCAD/EDA framing (Slide 6’s idea) to Slide 1; do NOT promote the 100% vs 12.5% chart in its current form to anywhere prominent. Before that chart can carry the front of the deck, it must be replaced with a powered, blinded, out-of-domain comparison (n ≥ 30, named comparator, named operator, public benchmark like FEP+ MERCK-set or PDBBind, scoring rubric stated). Until then, the chart belongs in the appendix as supporting anecdote, not as the cover claim. Narrative’s instinct to lead with a concrete proof point is correct; the specific proof point currently available isn’t strong enough to bear that weight.

3.2 Is the bioTCAD/EDA frame brilliant or reckless?

  • Narrative: “the cleverest pitch frame in the deck. Belongs on Slide 1.”
  • Devils: “dangerously optimistic. Invokes Synopsys/Cadence/Schrödinger directly — Schrödinger has been doing physics-grounded, parameter-validated drug simulation for 30+ years, is publicly traded, and still trades at a discount. A seed deck that picks the boldest possible analogy and then ducks every named incumbent reads as either unaware or evasive.”
  • Numbers: intellectually correct, but “bioTCAD will be as widely used as semiconductor EDA: a requirement for every launch” is overreach against n=8 evidence.

Reconciled recommendation. Keep the frame; earn it. The frame is the deck’s strongest IP — drop it and Erudio becomes another “AI + biology platform.” But the frame currently writes checks the evidence cannot cash. Two edits:

  • Name Schrödinger explicitly on Slide 6 in a single-line “How is bioTCAD different from FEP+/AlphaFold3/[other]?” Two specific differentiators (e.g., measurement-anchored parameter fitting from Erudio’s own chip; out-of-distribution behavior in novel chemical space) replace the implicit “we’re 8x better than the leader” claim with a defensible “we are differentiated on X and Y.”
  • Soften the universality claim from “as widely used as EDA: a requirement for every launch” to something earned: “a measurement-anchored simulator for medicinal-chemistry teams making novel modalities.” Save the “EDA universal-tool” framing for the Series A.

3.3 Which wedge to keep, and how harshly to cut?

  • Narrative: “pick one. Biodefense is the strongest narrative (field-deployable, fast actionable decision, government dollars). Pharma is the strongest frame. Either commit.”
  • Devils: “drop biodefense and clinical diagnostics. Pitch drug development only — bioTCAD as a paid SaaS/services tool for pharma medicinal-chemistry teams. Come back when the deck has one buyer.”
  • Narrative also offered a counter-pre-emption: “for SBIR / In-Q-Tel / strategic-defense audiences, three-market dual-use breadth is a feature, not a bug. If the audience is government rather than VC, my muddled-audience critique is mis-aimed.”

Reconciled recommendation. Two-deck strategy. The fundamental disagreement is audience-dependent: a defense-strategic buyer reads dual-use as strength; a tier-1 VC reads it as lack of focus.

  • For a VC seed round: use Devils’ framing — drug-development-only deck, single buyer (pharma medchem head), single product (bioTCAD-as-service or VSA-instrument-with-bioTCAD-bundle), single milestone path. Demote biodefense + clinical to a one-line “platform optionality” mention in the appendix.
  • For SBIR / In-Q-Tel / DoD / BARDA capability briefings: retain the three-market deck — biodefense leads, the rest is dual-use.

Erudio is currently using the second deck to try to do the first deck’s job. That is the source of the muddle every reviewer flagged.

3.4 Slide 5 (cytokine immunoassay) vs Slide 7 (peptide validation) — which is the more defensible hero?

  • Narrative: Slide 7 is “the killer.” Slide 5 is “a results poster.”
  • Devils (concession section): “Slide 5 is more interesting than Slide 7. Force-spectroscopy error removal on 6-plex cytokine speaks to a known assay pain point (multiplex crosstalk) that pharma actually pays to solve — a more believable platform claim than the peptide chart.”
  • Numbers: Slide 5 is concrete but under-quantified (cytokine identities, n, CV%, LoD missing). Slide 7 is dramatic but statistically thin.

Reconciled recommendation. Slide 5 is the more defensible hero; Slide 7 is the more dramatic one. For an investor who knows assay platforms, Slide 5’s claim (eliminated multiplex cross-reactivity at 6-plex, with a path to N-plex) is closer to a fundable product than Slide 7’s binding-prediction anecdote. The deck should reframe both:

  • Slide 5 reframe. Lead with the problem: “multiplex immunoassays plateau at 6-plex because cross-reactivity dominates above that.” Then the result: clean 6-plex, sourced LoD and CV% per analyte, head-to-head vs Luminex/MSD. Then the implication: path to 20-plex / 50-plex.
  • Slide 7 reframe. Already covered in §3.1. Move to appendix until n ≥ 30 blinded benchmark is available.

A defensible re-ordered story is: Problem (multiplex/in-silico unreliability cost pharma X) → Approach (measurement-anchored hybrid physics+AI = bioTCAD) → Proof Point 1 (Slide 5, quantified) → Proof Point 2 (Slide 7, appendix or right-sized) → Team → Ask.

4. Combined Slide-by-Slide Action List

Slide Current title Combined panel verdict Action
1 Erudio Bio — Data Driven AI for Effective Medicine Generic title card; wastes the deck’s most valuable slide. Replace with bioTCAD/EDA frame + 1 specific differentiator. Headline must be company-specific, not category-generic.
2 Erudio Technology Stack Architecture diagram before problem framing. Replace with quantified Problem slide. Move stack to slide 4 or appendix.
3 Multifaceted Data Platform: VSA Real assets (Stanford, 21 patents, chip), but “error-free” is an overclaim. Soften “error-free” → “error-suppressed.” Cite 21-patent table (USPTO #s, jurisdiction, lead-claim summary). Anchor “10 years” to a founding date.
4 Erudio VSA and DR-IN-A-BOX Naming sprawl begins (VSA, DR-IN-A-BOX, AI DR-IN-A-BOX, bioTCAD). Pick one product naming hierarchy. Make explicit what one customer buys.
5 Validation of Force Spectroscopy Error Removal on 6-plex Cytokine Immunoassay More defensible than Slide 7 but poorly framed. Reframe as Problem → Result → Implication. Quantify: cytokine identities, n, replicates, CV%, LoD, head-to-head vs Luminex/MSD.
6 bioTCAD: Hybrid AI & Physics-based Drug Development The deck’s strongest idea, buried. Promote the frame to Slide 1. Add competitive-differentiation lines (Schrödinger FEP+, AlphaFold3, etc.). Soften “requirement for every launch.”
7 Head-to-Head bioTCAD Validation: 100% vs 12.5% Dramatic visual; statistically fragile (n=8, anchor in train set, comparator unnamed). Move to appendix until n ≥ 30 blinded out-of-domain benchmark is run with a named comparator and stated rubric.
8 Erudio’s Technology Breakthrough — The Secret Sauce Synopsys/Samsung/SK Hynix logos imply partnership but are almost certainly founder employers. “Secret Sauce” tone off for defense/dx buyers. Label logos explicitly (“founder pedigree”). Replace “Secret Sauce” with neutral title. Tie founder credentials to specific capabilities, not adjectives.
9 Erudio Application: Drug Development Third architecture diagram. No customer, no pilot, no price, no workflow a pharma scientist would recognize. Replace with a real GTM slide: named target customer (Head of Computational Chemistry @ top-20 pharma), pricing model, 12-month pilot plan, 3 named target pharma partners.
10 Erudio Application: Biodefense Visually the strongest slide. Audience pivot is jarring. “Dozens, hundreds, thousands” is aspirational not capability. For the VC seed deck: remove. For a DoD/BARDA capability brief: lead with this slide and re-anchor SBIR/CRADA/program-of-record references.
11 Erudio Application: Clinical Diagnostic “20% of world’s economy” overstates by ~2× (actual ~10% global, ~17% US). SNUH/Keimyung “initial interests” not quantified. For VC seed deck: remove. If retained: replace TAM line with WHO GHED + CMS NHE figures; quantify hospital relationships (LOI, date, dollar amount).
12 Team / Advisory Board Deck’s second-strongest slide. But deck ends here — no ask, no use of funds, no roadmap. Keep. Add a closing slide 13 with: round size, valuation expectation, use of funds (3–5 buckets, %), runway, 12/18/24-month milestones. Disclose Tim Germann’s Carterra CCO role if Erudio’s chip is in any way adjacent to LSA.

5. Prioritized Fix List

Numbered roughly by impact-per-effort. The first five are unanimous; the rest reflect synthesis.

  1. Add the Ask slide. Round size, pre-money, use of funds, runway, 12/18/24-month milestones. Cost: one slide. Impact: turns the deck from a capability brochure into a fundable memo.
  2. Pick a single wedge for the VC version. Drug development (Devils’ choice; most consistent with bioTCAD framing). Remove Slides 10 and 11 from the VC deck. Build a separate one-pager / brief for biodefense.
  3. Rewrite Slide 1. Replace generic tagline with bioTCAD/EDA frame + one concrete differentiator. Lead the deck with the strongest idea, not a logo.
  4. Insert a quantified Problem slide as Slide 2. Replace the current architecture diagram. State the pain in pharma medchem dollars / months / failure rates.
  5. Add a competitive landscape slide. Name Schrödinger, Recursion, Isomorphic Labs, Iambic. Show where bioTCAD sits and what it does that they don’t (measurement-anchored parameter fitting from in-house chip biology).
  6. Replace the Slide 7 chart with a powered benchmark. n ≥ 30, blinded, out-of-domain, named comparator (Schrödinger FEP+ in expert configuration), public benchmark set, stated rubric. Until then, demote to appendix.
  7. Quantify Slide 5. Cytokine identities, n, replicates, CV%, LoD per analyte, dynamic range, head-to-head vs Luminex/MSD. This is your most defensible product claim — invest in making it bulletproof.
  8. Anchor the IP claim. “21 issued patents” → table of 5–10 lead patents with USPTO #s, jurisdictions, issue dates, one-line claims. Disclose Stanford license terms (exclusive? field-limited? royalty? sublicense rights?).
  9. Label the Synopsys/Samsung/SK Hynix logos as founder employers (or remove them).
  10. Quantify the hospital traction. SNUH and Keimyung as named LOIs/pilots with dates and dollar amounts, or remove. Same for Synopsys-Samsung-SK Hynix.
  11. Soften absolute language. “Error-free” → “error-suppressed.” “Requirement for every launch” → “measurement-anchored simulator.” “Massively multiplexed” → state demonstrated and target plex counts. “20% of world’s economy” → sourced WHO/CMS figures.
  12. Add a commercial hire to the team slide (or be explicit about the search). A former Head of Computational Chemistry at a top-20 pharma, or a SaaS-for-bio CRO operator, at founder-level equity. Cola as advisor is not a CCO.
  13. Disclose the Carterra conflict. Tim Germann is CCO of Carterra Bio, an adjacent biophysics company. Either explain the differentiation explicitly or move him off the formal advisory.

6. Path to “Investable”

What the panel collectively believes Erudio needs in the next 6 months to convert a “pass” into a “soft yes”:

  • One wedge picked, named in the deck, and selling. Three paying pharma pilots in the $50K–$250K range. LOIs count for one meeting; signed POs change the conversation.
  • A defensible hero benchmark. n=30 blinded out-of-domain head-to-head against Schrödinger FEP+ (in expert configuration, ideally run by a third party or with the protocol pre-registered). Beat them on a documented subset of chemistries; tie on the rest.
  • A commercial hire on the cap table. Customer-side credibility at founder level — not at advisor level.
  • An ask slide with milestones. Round, runway, use-of-funds, what the next 18 months produce, what triggers the Series A.
  • A 1-line patent moat statement. What the top 5 independent claims fence, and how Erudio’s biophysics chip is differentiated from Carterra’s LSA.

The underlying scientific idea — measurement-anchored, physics-grounded, AI-augmented simulation, with the measurement instrument built in-house — is the right idea, and the team’s chip + biophysics + optimization-trained AI composition is genuinely rare. The deck is currently doing that idea a disservice by burying it under three architecture diagrams, three application slides for three different audiences, and a hero statistic that’s statistically indistinguishable from a coin flip at the demonstrated sample size. Fix the deck; the company is more credible than the deck makes it look.

Appendix A: Reviewer hand-off notes (preserved verbatim, would have been SendMessage exchanges)

The sub-agent sandbox prevented direct peer-to-peer messaging, but each reviewer drafted explicit hand-offs to the other two. These are captured here so the user can see where each reviewer pre-emptively conceded or pushed back on the others’ likely arguments.

Narrative → Numbers. “My narrative review is final. Two places I’m exposed if your numbers come back ugly: (1) I declared Slide 7 the deck’s killer slide on the strength of ‘100% vs 12.5% / 8 designs’ — if n=8 collapses or the comparison setup is cherry-picked, my entire top-of-deck reordering breaks; (2) Slide 6’s ‘21 patents / 10 years’ and ‘Synopsys/Samsung/SK Hynix’ do load-bearing credibility work — if any are misleading, my Slide-8 critique (under-explained logos) becomes a much harsher ‘remove these’ call.” → Numbers confirmed both exposures. The reconciled recommendation in §3.1 routes around the Slide 7 problem by promoting the bioTCAD frame (Slide 6’s idea) rather than the chart.

Narrative → Devils. “Push me on recommendation #3 (pick one wedge). Plausible counter: for SBIR / In-Q-Tel / strategic-defense audiences, three-market dual-use breadth is a feature, not a bug — and ‘GWOT to Great Power Competition’ may be intentionally aimed at a non-VC buyer. If that’s the actual audience, my muddled-audience critique is mis-aimed and the deck is more coherent than I’m giving it credit for.” → Devils held the harder line (“drop biodefense and clinical from the seed pitch”). Reconciled in §3.3: two-deck strategy.

Devils → Narrative. “Three structural problems hurt the story even more than the science: (1) Slides 9–11 read as three separate companies; (2) Slide 6’s ‘bioTCAD = EDA’ promise is a check the rest of the deck does not cash, because no incumbent (Schrödinger, Recursion, Iambic, Isomorphic) is named; (3) Slide 11’s ‘20% of the world’s economy’ is a TAM-substitute that signals the company doesn’t know who its first 10 customers are.”

Devils → Numbers. “Two figures need scrubbing. (1) Slide 7’s ‘100% vs 12.5%’ — n=8, one chemical series, one of the eight is the anchor used to fit the force field. Effective sample size is 7, and ‘12.5% legacy’ likely reflects an unoptimized run. (2) Slide 3’s ‘10 years development, 21 issued patents’ — confirm issuance vs. application, jurisdictions, assignee. Also confirm whether Slide 8’s Synopsys/Samsung/SK hynix logos represent customer relationships, employer affiliations, or aspirational targets — that distinction is material.” → Numbers’ audit table confirmed all three flags.

Numbers → Narrative. “Numbers issues that hurt narrative credibility: ‘Error-free’ (slide 3) contradicts ‘error removal’ (slide 5); ‘bioTCAD will be a requirement for every launch’ (slide 6) lands as overreach against n=8 evidence; ‘Initial interests’ presented as traction is unquantified; Synopsys/Samsung/SK Hynix logos imply partnership; appear to be former employers.”

Numbers → Devils. “Strongest bull data: Slide 5’s force-spectroscopy + Bayesian filtering on 6-plex cytokine is falsifiable and demonstrated; founder pedigree (Stanford + TCAD industry) is real; Davis/Snyder/Greenleaf advisory is top-tier. Strongest bear data: n=8 with overlapping confidence intervals (8/8 CI [63–100%] vs 1/8 CI [2–47%] overlap); no ask, no traction, no regulatory plan, no sourced TAM; pattern of unhedged absolutes signals founder overclaiming; ‘$15B+ exits’ advisor attribution unverifiable.” → Devils’ concessions section accepted Slide 5 as more interesting than Slide 7; reconciled in §3.4.