Academic Commercialization Assessment Agent

Turn any research paper or topic into a commercialization readiness report in minutes — six AI agents gather academic, patent, and market evidence, then produce a scored report with verifiable citations.

English | 中文

English

A multi-agent system built on CrewAI that evaluates the commercialization readiness of academic research.

Input a research direction or paper topic. Six specialized AI agents automatically gather evidence from academic literature, patent databases, and market intelligence sources, then produce a structured commercialization assessment report with verifiable citations and a quantitative scorecard.

Screenshots

What's different from the CrewAI starter template

Agent architecture

Agent 1: Academic Literature Analyst
         Sources: OpenAlex / Semantic Scholar papers (pre-validated in Step 0)
         Output:  Structured EvidenceReport JSON — maturity, breakthroughs, citations (A1/A2/…)

Agent 2: Patent Landscape Analyst
         Sources: Google Patents / WIPO records (Serper search + URL validation)
         Output:  Structured EvidenceReport JSON — holders, white spaces (P1/P2/…)

Agent 3: Market & Competitive Intelligence Analyst
         Sources: Domain-allowlisted market reports (Serper search)
         Output:  Structured EvidenceReport JSON — players, target industries, opportunities (M1/M2/…)

Agent 4: Technology Commercialization Report Writer
         Tools:   None (uses Agents 1–3 output as context only)
         Output:  Markdown draft with inline citations [A1][P2][M3] and References block
         Guard:   Section structure + citation integrity; auto-retries up to 2×

Agent 5: Report Reviewer
         Tools:   None (uses Agent 4 draft as input)
         Rules:   6 rules — citation integrity, unsupported numeric claims, overconfident
                  language, patent legal framing, evidence consistency, TRL label consistency
         Output:  Corrected final report; Reviewer Notes saved separately (only actual changes logged)

Agent 6: Commercialization Readiness Scorer
         Tools:   None (reads Tasks 1–3 evidence JSON directly, independent of the report)
         Output:  CommercializationScore JSON — TRL / MRL / Patent / Market / Evidence confidence
         Guard:   JSON format validation + hallucinated source ID check + weighted formula
                  correction; auto-retries up to 2×

Agents 1–3 run in parallel (async_execution=True), reducing total pipeline time.

Execution flow

Step 0  Source collection & validation (subprocess, deterministic)
        Academic: OpenAlex Works API (filter=title.search, sorted by citation count)
                  → Semantic Scholar supplement (when OpenAlex count is below target)
                  → DOI deduplication; summaries < 100 chars auto-rejected
                  → Concurrent Crossref citation-count backfill (ThreadPoolExecutor)
        Patent:   Serper (3-attempt retry with exponential backoff) → Google Patents / WIPO;
                  URL reachability verified; patent hosts short-circuited
        Market:   Serper + domain allowlist (30+ approved institutions); low-quality sites removed
        Metadata: Crossref API for DOI, journal name, publication date
        Output:   validated_sources.json + status.json passed to subprocess pipeline

Steps 1–3  Agents 1/2/3 — Academic / Patent / Market analysis  (parallel)
Step 4     Agent 4 — Comprehensive report writing  (guardrail validates citations)
Step 5     Agent 5 — Quality review  (Reviewer Notes saved separately)
Step 6     Agent 6 — Quantitative scoring  (independent of report; formula auto-corrected)

The pipeline runs in a subprocess (pipeline_worker.py) so the Gradio UI can cancel it immediately via proc.terminate().

Report structure

# Academic Commercialization Assessment: <research_topic>
## Executive Summary
## 1. Technology Overview & Maturity
## 2. Patent Landscape & White Spaces
## 3. Target Industries & Use Cases
## 4. Competitive Landscape
## 5. Commercialization Opportunities & Recommendations
## Evidence Limitations
## References
    *Reference codes: A = Academic paper · P = Patent · M = Market/industry source*
    [A1] … [P1] … [M1] …

Multilingual support: Language is auto-detected from the topic string. Reports in Simplified/Traditional Chinese, Japanese, Korean, German, French, and 6 more languages are fully localized — section headings, citation legend, and patent disclaimers all adapt automatically.

The scorecard (commercialization_scores.json) additionally contains: TRL score, patent strength, market accessibility, evidence confidence, overall score, key risks, and key opportunities.

See the examples/ folder for three complete real reports across different industries.

Sample output

Overall Score: 41.3 / 100 · Biotech profile · TRL 3.3/9 · MRL 2.0/10 · Patent 4.0/5 · Market 2.0/5 · Evidence 3.0/5

Executive Summary

This assessment evaluates single-dose LNP-ABE8e base editing therapy for hemophilia A targeting the F8 Arg2038Cys mutation. The technology demonstrates strong preclinical proof-of-concept with durable factor VIII restoration (83.7 ± 9.1% at 52 weeks, no detectable immune response [A1]) but remains at an early readiness stage with significant clinical development milestones ahead.

1. Technology Overview & Maturity

In vivo adenine base editing via LNP delivery achieved 72 ± 8% on-target A-to-G correction in bulk liver tissue at 4 weeks, with FVIII activity restoration confirmed durable to 52 weeks [A1]. Tail-clip bleeding time normalised to 2.8 ± 0.4 min (wild-type: 2.6 ± 0.3 min) [A1][A2]. No off-target editing or immune response was detected.

TRL 3.3 / 9 — Active R&D with in vivo animal model proof-of-concept; no IND filing or human data available.

2. Patent Landscape & White Spaces

LNP-mediated liver delivery for in vivo base editing is actively patented by Beam Therapeutics, Intellia Therapeutics, and Precision BioSciences [P2][P4]. The specific F8 Arg2038Cys correction approach may represent a differentiated white space, but freedom-to-operate analysis is required before clinical or commercial use [P1][P3].

Evidence Limitations

• Academic: All efficacy data from a single murine model; primate or human data absent [A1]
• Patent: USPTO PatentsView API retired — patent search via Google Patents / WIPO only [P1–P6]
• Market: Hemophilia A gene therapy market projections vary widely ($2.9B–$4.3B by 2032) [M1][M3]

References

• [A1] Kim et al. (2025). In Vivo CRISPR-Cas9 Base Editing Achieves Durable Factor VIII Restoration… https://doi.org/10.1038/s41587-025-02314-8
• [P1] ENDONUCLEASE FOR TARGETING BLOOD COAGULATION FACTOR VIII — Google Patents
• [M1] Hemophilia Gene Therapy Market Report, Market Size and Revenue… — Market Research Future

Quick start

1. Install dependencies

uv sync

2. Configure environment variables

Copy .env.example to .env and fill in your keys:

cp .env.example .env

LLM — pick one of:

Also required:

Optional: