Hermes Nexus - Enhanced Product Vision

Hermes Nexus is an AI-native deal execution platform that transforms fragmented procurement, contracting, compliance, and enablement workflows into a single evidence-backed workspace.

Comment

Inspiration

We were inspired by the universal frustration of fragmented deal execution. In our own experiences and through countless conversations with procurement, legal, and operations teams, we saw a recurring pattern: critical vendor information was scattered across emails, PDFs, call recordings, and spreadsheets, leading to risky blind spots, compliance scrambles, and lost institutional knowledge. Existing tools were either passive repositories (like contract management systems) or isolated point solutions that didn't talk to each other. We envisioned a platform that could act as an active reasoning partner—one that could connect promises across proposals, calls, and contracts to prevent "verbal deal drift," auto-generate compliance evidence, and turn every negotiation into a reusable lesson. The Gemini 3 Hackathon's focus on "Marathon Agents" and long-running workflows cemented our vision: to build an AI-native workspace that orchestrates the entire vendor lifecycle, from evaluation to renewal, with verifiable evidence at its core.

What it does

Hermes Nexus is an AI-native deal execution platform that transforms chaotic procurement, contracting, and compliance workflows into a single, evidence-backed workspace. It actively ingests proposals, sales call recordings, contracts, and internal meetings to extract key commitments, detect contradictions, assess risks, and generate actionable outputs—all with traceable citations to the source material.

Key capabilities include:

  • Procurement Evaluator: Upload vendor proposals for AI-powered, apples-to-apples comparison with requirement-level scoring.
  • Contract Redliner: Analyze contracts to flag risks, suggest redlines, and detect mismatches against earlier promises.
  • Sales Call Promise Tracker: Extract and timestamp vendor commitments from recordings to prevent "he-said-she-said" disputes.
  • Compliance Evidence Pack Builder: Auto-link policies and contract clauses to security controls (SOC2/ISO) and pre-fill questionnaires.
  • Form Filler: Automatically populate repetitive forms (W-9s, onboarding sheets) from a canonical profile, with consistency checks.
  • Training Library: Automatically segment recordings into searchable clips that surface just-in-time during active deal work.

How we built it

We built Hermes Nexus as a full-stack application architected around a hybrid knowledge graph, designed to orchestrate complex, multi-step AI workflows.

  • Frontend: A modern React/TypeScript application with a workspace-style UI, providing a unified "Deal Room" for all vendor interactions.
  • Backend & Orchestration: Python/FastAPI backend managing the core application logic and multi-step AI agent pipelines. We leveraged Gemini 3 Pro for its long-context reasoning and multimodal capabilities (processing text, audio, and video) to power deep extraction, contradiction detection, and summarization across documents and recordings.
  • Knowledge Graph & Data Layer: Neo4j stores the interconnected "Evidence & Commitments Graph," linking Sources, Facts, Risks, and Tasks. This enables powerful queries like "Find all promises that contradict this contract clause."
  • AI/ML Stack:
    • Gemini 3 Pro for high-level reasoning, extraction, and summarization.
    • Groq API (with Whisper) for fast, real-time transcription of audio/video.
    • Custom prompts and multi-pass verification loops to ensure extraction accuracy and minimize hallucinations.
  • Integrations & Storage: Integrated with cloud storage (for document upload), and designed future integrations for e-signature (DocuSign) and communication platforms (Slack, Gmail).

Challenges we ran into

  • Ensuring Citation Accuracy: Building a system where every AI-generated claim is backed by a verifiable source (page number, timestamp) required sophisticated chunking, embedding, and retrieval-augmented generation (RAG) techniques, coupled with post-extraction verification loops.
  • Graph Complexity: Designing a knowledge graph schema flexible enough to connect diverse entities—from contract clauses and sales promises to compliance controls and training clips—while maintaining performance was a significant modeling challenge.
  • Contradiction Detection Across Modalities: Teaching the AI to identify semantic conflicts (e.g., a verbal promise of "month-to-month" vs. a contract clause for "annual commitment") across different document types and unstructured audio required nuanced prompt engineering and cross-graph querying.
  • Orchestrating Long-Running Workflows: Implementing the "Marathon Agent" pattern—where a deal state persists over days or weeks, triggering different AI and human tasks—required robust state management and event-driven architecture.
  • Balancing Autonomy & Trust: Determining the right "human-in-the-loop" gates was critical. We had to design the UI to build user trust by transparently showing citations and confidence scores, while still automating the bulk of tedious work.

Accomplishments that we're proud of

  1. Building a Fully Functional, Integrated Prototype: In a short timeframe, we created a cohesive platform that demonstrates the core vision—from proposal upload to risk-flagged contract and a filled compliance form—all within a single, intuitive workspace.
  2. The Evidence Graph: Successfully implementing the core innovation: a working knowledge graph that actively links extracted facts to their sources and flags contradictions, moving far beyond simple document storage.
  3. Cross-Module Intelligence: Creating genuine synergy between features (e.g., a promise from a sales call automatically triggering a contradiction warning in the Contract Redliner) proves the value of a unified system over point solutions.
  4. Practical AI Application: Developing practical, high-ROI AI features that solve acute pain points (like auto-filling security questionnaires with citations) rather than just demo-grade "magic."
  5. User-Centric Design: Crafting a complex product that remains approachable for non-technical users in procurement, legal, and operations, making enterprise-grade rigor accessible to smaller teams.

What we learned

  • Evidence is Everything: For users to trust AI with critical business decisions, traceability is non-negotiable. The most powerful feature isn't the AI's answer—it's the footnote showing where it came from.
  • Integration Creates Compound Value: The whole product becomes exponentially more valuable than the sum of its parts when modules share data. A promise extracted in one module must be usable as a constraint in another.
  • The "Last Mile" is Crucial: AI can draft a perfect redline or fill a form, but the final step—generating the email to send to the vendor, or the PDF to submit—is what delivers complete time savings.
  • Institutional Knowledge as a Byproduct: The most sustainable training content isn't created intentionally; it's automatically captured from real work (like a deal post-mortem) and surfaced in context.
  • SMBs Need Enterprise Tools Too: Small and mid-size teams face the same vendor, contract, and compliance complexities as large enterprises but lack the dedicated personnel. They are a hungry, underserved market for sophisticated tools made simple.

What's next for Hermes Nexus - Enhanced Product Vision

Our immediate next steps focus on transforming our hackathon prototype into a robust, user-ready product and beginning our go-to-market journey.

  1. Closed Beta & Pilot Program: Launch an invite-only beta with 20-30 design partner teams from our target market (growth-stage SaaS, professional services) to gather intensive feedback and validate core workflows.
  2. Enhanced AI & Reliability: Implement multi-model voting (e.g., Gemini + Claude) for critical extractions, expand our library of clause playbooks and compliance frameworks, and harden our verification pipelines.
  3. Freemium Launch & Self-Serve Onboarding: Develop a public freemium tier (3 active deals, limited uploads) to drive organic user acquisition, complemented by in-app tutorials and interactive walkthroughs.
  4. Key Integrations: Build native integrations for e-signature (DocuSign), cloud storage (Google Drive, SharePoint), and communication (Slack, Microsoft Teams) to fit seamlessly into existing workflows.
  5. Monetization & GTM: Officially launch our paid "Pro" and "Team" tiers, begin content marketing targeting procurement and ops leaders, and establish a partnership channel with compliance consultants.
  6. Platform Expansion: Begin developing the marketplace for third-party clause playbooks and industry-specific compliance packs, and explore API partnerships with procurement/P2P software platforms.

Built With

Share this project:

Updates