LexBridge: Cross-Lingual M&A Due Diligence Intelligence

Everything runs locally. No cloud. No API keys. No data leaves your machine.

A fully local, multi-agentic system for cross-lingual M&A due diligence analysis -- powered by Microsoft Harrier multilingual embeddings, Google Gemma 4 E2B LLM (via llama.cpp), Qdrant vector search, FalkorDB knowledge graph, and LangGraph multi-agent orchestration.

Type "Zeige mir alle Dokumente ueber Produkthaftungsrisiken" (German) and find English Monsanto lawsuit filings, EPA correspondence, AND German Bayer risk memos -- with zero translation.

Screenshots

LLM-Powered Cross-Lingual Analysis

Gemma 4 E2B analyzes retrieved documents across languages and synthesizes findings in the query language -- all running locally on Metal GPU via llama-server.

Cross-Lingual Document Retrieval

A German query ("Produkthaftungsrisiken") retrieves both German Bayer internal documents AND English Monsanto lawsuit filings, insurance policies, and SEC disclosures -- ranked by semantic similarity with zero translation.

Entity Relationship Graph

Knowledge graph showing Monsanto -> Roundup -> Glyphosate -> Lawsuits -> Courts, with color-coded nodes for Companies (blue), Products (green), Legal Claims (red), Regulatory Bodies (orange), and Chemicals (purple).

Risk Dashboard

48 documents, 235 graph nodes, 264 relationships across 2 languages. Documents by language and type, risk factors ranked by frequency.

The Bayer-Monsanto Story: Why This Demo Exists

In 2018, Bayer AG (Germany) acquired Monsanto (USA) for $63 billion. Bayer's German-speaking legal team catastrophically underestimated Monsanto's US litigation exposure from Roundup (glyphosate) product liability lawsuits:

• 67,000+ lawsuits alleging Roundup causes non-Hodgkin lymphoma
• $10B+ already paid in settlements
• $50B+ in total shareholder value destruction -- the most expensive due diligence failure in M&A history

The liability was scattered across thousands of English-language US court filings, EPA regulatory correspondence, and internal Monsanto documents. Bayer's German legal team either couldn't efficiently review this volume of English-language documents or underestimated the patterns hiding in them.

The demo thesis: "If Bayer had used cross-lingual contract analysis to search ALL of Monsanto's US regulatory filings, litigation history, and product liability clauses -- from their German-speaking legal team -- they would have seen the true scope of Roundup liability before paying $63 billion."

What It Does

• Cross-lingual document search -- German lawyer searches "Produkthaftung" (product liability) and finds all relevant English documents about Roundup litigation
• Risk assessment across jurisdictions -- Automatically flag liability exposure by comparing US tort law documents with German regulatory standards
• Entity relationship mapping -- Knowledge graph showing connections: Monsanto -> Roundup -> Glyphosate -> Lawsuits -> Courts -> Settlement amounts
• Deal risk dashboard -- Risk factors by category and severity, litigation exposure charts, document coverage analysis
• Due diligence gap analysis -- "68% of English-language US litigation filings have no corresponding reference in any German-language due diligence report"
• Clause comparison -- Cross-lingual similarity matrix comparing indemnification provisions across merger documents

Tech Stack (100% Local, No Cloud Required)

Component	Technology	Role
Embeddings	Harrier-OSS-v1-0.6B (MPS GPU)	94-language multilingual embeddings, 1024-dim, 32K context
LLM	Gemma 4 E2B-it Q4_K_M via llama-server (Metal GPU)	Intent classification, Cypher generation, risk analysis synthesis
Vector DB	Qdrant (embedded, no server)	Semantic similarity search over document embeddings
Graph DB	FalkorDBLite (embedded)	Litigation chains, entity relationships via Cypher
Agents	LangGraph (StateGraph)	Multi-agent orchestration with conditional routing
UI	Streamlit	Interactive web dashboard with search, graph viz, risk dashboard

Runs entirely on a MacBook with Apple Silicon (M1/M2/M3/M4) and 8GB RAM.

Quick Start

Requirements

• macOS with Apple Silicon (M1/M2/M3/M4)
• Python 3.12 (must be arm64 -- see Troubleshooting if unsure)
• Xcode Command Line Tools: xcode-select --install
• Homebrew (arm64, installed at /opt/homebrew): needed to install llama.cpp with Metal GPU
• ~8GB free disk space (dependencies ~2GB + Harrier model ~1.5GB + Gemma 4 model ~3.2GB)

Setup (~10 minutes)

# 1. Clone (~1s)
git clone https://github.com/pankajarm/LexBridge.git
cd LexBridge

# 2. Install llama.cpp for the LLM server (~30s)
#    Gemma 4 requires build b8636+. Use --HEAD if stable is too old.
brew install llama.cpp --HEAD

# 3. Create a native arm64 virtual environment (~2s)
#    IMPORTANT: The venv MUST use arm64 Python, not x86_64/Rosetta.
#    PyTorch >= 2.4 only ships macOS arm64 wheels.
python3.12 -m venv .venv
source .venv/bin/activate

# Verify architecture (must print arm64, NOT x86_64):
python -c "import platform; print(platform.machine())"

# 4. Install Python dependencies (~2 min)
pip install -e .

# 5. Copy environment config
cp .env.example .env

# 6. Download Harrier embedding model (~2 min, ~1.5GB)
python scripts/download_models.py

# 7. Start Gemma 4 LLM server in a separate terminal (~3 min first run)
#    Auto-downloads the 3.2GB model on first run, then starts serving.
#    Keep this running in the background.
llama-server -hf lmstudio-community/gemma-4-E2B-it-GGUF -ngl 99 -c 4096

# 8. Generate sample data + build databases (~1 min)
#    (Back in your original terminal with the venv activated)
python scripts/setup_databases.py

# 9. Launch web UI
streamlit run src/ui/app.py

Open http://localhost:8501 in your browser.

Setup Time Breakdown

Step	Command	Time	Notes
1. Clone	git clone	~1s
2. Install llama.cpp	brew install llama.cpp --HEAD	~30s	Needs Xcode CLT for HEAD build
3. Create venv	python3.12 -m venv .venv	~2s	Must be arm64 Python
4. Install Python deps	pip install -e .	~2 min	PyTorch, Transformers, Streamlit, etc.
5. Config	cp .env.example .env	instant
6. Download Harrier	python scripts/download_models.py	~2 min	Harrier 1.5GB embedding model
7. Start LLM server	llama-server -hf lmstudio-community/gemma-4-E2B-it-GGUF -ngl 99 -c 4096	~3 min	Auto-downloads 3.2GB on first run
8. Setup databases	python scripts/setup_databases.py	~1 min	48 documents, 235 graph nodes
9. Launch	streamlit run src/ui/app.py	~10s	First load warms up models
Total		~10 min	Tested on M3 8GB

CLI Mode

python main.py

Memory Usage (8GB Mac)

Component	RAM
Harrier embeddings (MPS)	~1.5 GB
Gemma 4 E2B Q4 via llama-server (Metal)	~3.2 GB
Qdrant + FalkorDB	~100 MB
Streamlit + Python	~200 MB
Total	~5.0 GB

Some swap usage is expected on 8GB machines when both models are loaded. Performance remains good thanks to unified memory and Metal GPU acceleration.

Architecture

                    +-------------------------------------+
                    |         Streamlit Web UI             |
                    +--------------+----------------------+
                                   |
                    +--------------v----------------------+
                    |     LangGraph Supervisor Agent       |
                    +--+-------+-------+------+----------+
                       |       |       |      |
          +------------v-+ +--v-----+ +v-----v--+ +-------------+
          |  Ingestion   | |Semantic| | Graph    | | Risk        |
          |  Agent       | |Search  | | Query    | | Analysis    |
          +------+-------+ +--+-----+ +--+------+ +------+------+
                 |            |          |               |
    +------------v------------v--+  +----v----+  +------v--------+
    |   Qdrant (embedded)       |  | FalkorDB|  | llama-server   |
    +---------------------------+  | (embed) |  | Gemma 4 (Metal)|
                 |                 +----+-----+  +------+---------+
    +------------v----------------------v---------------+
    |        Harrier-OSS-v1-0.6B (MPS GPU)              |
    +---------------------------------------------------+

Agent Flow

User Query
    |
    v
[Supervisor] -- classifies intent (LLM or keyword fallback)
    |
    +-- document_search -----> [Semantic Search] --> [Synthesizer]
    +-- risk_assessment -----> [Semantic Search] --> [Graph Query] --> [Risk Analysis] --> [Synthesizer]
    +-- entity_lookup -------> [Graph Query] --> [Synthesizer]
    +-- clause_comparison ---> [Semantic Search] --> [Risk Analysis] --> [Synthesizer]
    +-- gap_analysis --------> [Semantic Search] --> [Graph Query] --> [Risk Analysis] --> [Synthesizer]

Graph Schema

(Company)-[:ACQUIRED {date, value}]->(Company)
(Company)-[:MANUFACTURES]->(Product)
(Product)-[:CONTAINS_CHEMICAL]->(Chemical)
(Chemical)-[:LINKED_TO]->(RiskFactor)
(Document)-[:MENTIONS]->(Company)
(Document)-[:DISCLOSES]->(RiskFactor)
(Document)-[:CONTAINS]->(ContractClause)
(Document)-[:SIMILAR_TO {score, cross_lingual}]->(Document)
(LegalClaim)-[:FILED_AGAINST]->(Company)
(LegalClaim)-[:TARGETS]->(Product)
(LegalClaim)-[:ALLEGES]->(RiskFactor)
(RegulatoryBody)-[:INVESTIGATES]->(Product)

Demo Queries

Try these in the UI:

1. Cross-lingual search (the hook): "Zeige mir alle Dokumente ueber Produkthaftungsrisiken" -- German query finds English Monsanto lawsuit filings AND German Bayer risk memos
2. Litigation exposure: "What is the total litigation exposure from glyphosate claims?" -- Graph traversal aggregates claim amounts
3. Clause comparison: "Compare indemnification clauses across merger documents" -- Similarity matrix reveals gaps between merger agreement caps and unlimited distributor indemnities
4. Gap analysis: "What due diligence gaps exist in the Bayer-Monsanto review?" -- Cross-lingual gap analysis identifies missing coverage
5. Entity lookup: "Show all entities connected to Roundup" -- Knowledge graph traversal
6. Risk assessment: "Find all Roundup product liability filings" -- Cross-jurisdictional risk aggregation

How Harrier Makes This Possible

Harrier-OSS-v1 maps text from 94 languages into a single shared 1024-dimensional vector space. This means:

• "Product liability" (English) and "Produkthaftung" (German) are geometrically close
• "Glyphosate" and "Glyphosat" map to nearby vectors
• "Indemnification" and "Freistellungsklausel" are semantic neighbors
• Cross-lingual search is just cosine similarity -- no translation pipeline needed

Sample Data

The demo includes 48 synthetic documents modeled after real M&A due diligence data room contents:

English (Monsanto/US side) -- 22 documents:

• Product liability lawsuit filings (Johnson v. Monsanto, Pilliod, Hardeman, MDL 2741, Dicamba)
• EPA regulatory correspondence and IARC classification
• Internal risk assessments (projecting 20,000-50,000 claims, $2-7B exposure)
• Settlement agreements ($78.5M bellwether, $10.9B global framework)
• Insurance policies ($750M tower vs $3-8B exposure -- massive gap)
• Indemnification clauses (unlimited distributor indemnity)
• SEC filings (10-K risk factors, 8-K merger announcement with MAE clause)
• Scientific studies (IARC evaluation, EPA assessment, independent meta-analysis, Monsanto-funded study)

German (Bayer side) -- 18 documents:

• Board meeting minutes ("Haftungsrisiko als beherrschbar eingestuft" -- liability rated as manageable)
• Due diligence reports (estimating only $100-240M exposure, "Worst Case" $1-2B)
• BaFin regulatory filings
• Legal memos comparing US/German product liability law (unlimited US liability vs EUR 85M German cap)
• Integration planning documents
• Shareholder communications
• Contract templates (German liability caps vs Monsanto's unlimited provisions)

Bilingual -- 8 documents:

• Merger agreement key terms (MAE excludes litigation under $3B, indemnification capped at $2B)
• Cross-border regulatory approvals (EU Commission, Bundeskartellamt, DOJ, CFIUS -- none reviewed product liability)

The "Aha" Moments

1. German query finds English smoking guns: Search "Produkthaftungsrisiken" and the top results include English-language Monsanto internal risk assessments projecting billions in liability -- documents the German DD team apparently never fully reviewed.

2. Indemnification gap exposed: The merger agreement caps Bayer's indemnification at $2B, but Monsanto's distributor contracts contain UNLIMITED product liability indemnification across 200+ agreements worldwide.

3. 68% documentation gap: 68% of English-language US litigation filings have no corresponding reference in any German-language due diligence report.

4. Insurance tower inadequacy: Total insurance coverage of $750M against projected exposure of $3-8B. The insurance policies themselves contain "known liability" exclusions that likely preclude Roundup coverage entirely.

5. Due diligence underestimation: Bayer's DD estimated worst-case at $1-2B. Monsanto's own internal projections (not disclosed to Bayer) estimated $2-7B. Actual cost: $10B+ and counting.

Troubleshooting

llama-server not found

Install llama.cpp via Homebrew (must be arm64 Homebrew at /opt/homebrew for Metal GPU):

brew install llama.cpp --HEAD

The --HEAD flag builds from the latest source (requires Xcode Command Line Tools). If that fails, try the stable bottle first:

brew install llama.cpp

If the stable version gives unknown model architecture: 'gemma4', the bottle is too old. Either retry with --HEAD or download a pre-built binary (build b8636+) from llama.cpp releases: Download the llama-*-bin-macos-arm64.tar.gz asset from the latest release, extract it, and run llama-server from the extracted directory.

LLM analysis shows mock responses

The LLM server isn't running. Start it in a separate terminal:

llama-server -hf lmstudio-community/gemma-4-E2B-it-GGUF -ngl 99 -c 4096

Wait until you see HTTP server listening before using the app. The app gracefully falls back to keyword-based mock responses when the server is unavailable.

No Metal GPU (slow inference)

If llama-server reports no usable GPU found, your llama.cpp was compiled without Metal support. This happens with x86_64 Homebrew under Rosetta. Fix:

# Install arm64 Homebrew (if you only have x86_64 at /usr/local)
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"

# Then install llama.cpp with arm64 brew
/opt/homebrew/bin/brew install llama.cpp

Python reports x86_64 instead of arm64

Your venv was created with an x86_64 Python (Rosetta). Recreate it:

rm -rf .venv
# Use a universal or arm64 Python binary:
arch -arm64 python3.12 -m venv .venv
source .venv/bin/activate

Related Projects

• MedBridge -- Same architecture applied to multilingual clinical trial intelligence (7 languages, drug interactions, cross-cultural pharmacovigilance)

License

MIT