Silver Tier — The Claim
Resume: “AWS Professional”
The system ingests the resume as a distinct source node. It records the text "AWS Professional" but tags it as unverified claim data.
Standard AI offers a "black box" probability. The Double-Sided Context Graph delivers Auditability by Design, validating the source behind every decision and driving toward a Pink Team decision.
The moment a new RFP is released, we need to identify win themes and compliance obligations immediately. In practice, the first days are spent simply hunting across Salesforce, SharePoint, and tribal memory just to assemble context.
Standard RAG automation fails because it lacks discernment. To a vector database, a 10-year-old resume is treated with the same weight as a signed contract. It cannot distinguish a "Claim" from "Evidence."
Ingests all documents equally. No concept of time or truth.
Retrieves chunks with high keyword similarity.
Synthesizes answer based on retrieved chunks.
"Candidate is a perfect match based on the requirements."
Mapping terms to Concept ID: 101
REJECTED
Evidence (Associate) contradicts Claim (Pro).
We reject the concept of static truth. By separating Data from Context, we validate the two variables that actually matter in procurement: Time and Provenance.
Amendments supersede original requirements. We track the active state of every rule.
AI-assisted normalization bridges the gap between different terms (e.g., "AWS Pro" ≈ "Solutions Architect") to prevent false negatives.
Claims are weighed by their evidentiary context. We prioritize corroborated records over unverified assertions by analyzing the full chain of custody.
Government procurements are chaotic. Requirements change via Amendments, and ambiguities are resolved via Q&A. We reify "Time" to treat the RFP not as a static file, but as a stream of evolving atoms.
"Senior Architect with Generic Cloud Certification"
"Does 'Generic' accept Azure?"
"No, must be AWS."
"Senior Architect with AWS Professional Certification"
Different documents use different words for the same concept. We use AI-assisted normalization to map messy language to a controlled concept graph, preventing "near-miss" hallucinations.
The RFP asks for a "Senior Cloud Architect." Your best candidate, Sarah, lists "Principal AWS Lead" on her resume. Different words, same job. A keyword search misses her entirely.
Our Semantic Bridge separates exact matches, directional implications, and related-but-not-sufficient concepts. "Principal AWS Lead" implies the capability of a "Senior Cloud Architect" in this domain, but the reverse is not true.
Resumes are claims, not facts. The context graph attaches provenance so verified sources carry greater evidentiary weight than unverified assertions, classifying claims into explicit trust tiers.
Silver Tier — The Claim
The system ingests the resume as a distinct source node. It records the text "AWS Professional" but tags it as unverified claim data.
Gold Tier — The Evidence
A recent statement of work is treated as verified evidence. When higher-trust evidence conflicts with lower-trust claims, the evaluation logic weights the evidence based on provenance.
In regulated industries like GovCon, the "why" is more important than the "what". A standard AI tells you "Sarah is a match." Our Context Graph provides the Traceability, Version Control, and Risk Mitigation required to defend that decision— driving directly toward a Pink Team decision.
Resume claims AWS Pro but verified source indicates AWS Associate.
"We rejected this candidate because verified certification (Source: Database) does not meet criteria." Every decision is linked to an immutable source record.
"We evaluated against Amendment 02, not the original RFP." The system never validates compliance against dead requirements.
Contradictions are surfaced, not suppressed. The system explicitly flags discrepancies to the Human Auditor before a risk can become a liability.
The challenge asked for an AI capability to assess bid viability, staffing, and win probability—The "Pink Team" MVP. The Double-Sided Context Graph is not just an abstract architecture; it maps 1-to-1 with the project deliverables.
Instead of a "vibe-based" prediction, the Graph calculates PWin (Probability of Win) by normalizing historical data—past submissions, wins, and project performances.
The "Sarah" scenario scaled up. We calculate a Confidence Score for each candidate based on evidentiary weight.
Comparative analysis against the Demand Plane.
Because the Context Graph is structured data, the final output is a document, not a chat message.
The Double-Sided Context Graph is more than a narrative—it is an architectural blueprint for safe, scalable AI. We have moved beyond simple prompt engineering to define the structures, responsibilities, and risk controls required to turn a conceptual idea into a deployable solution.
By decoupling the Context Graph reasoning engine from underlying data sources, the solution can be seamlessly ported between different organizational tenants. This ensures strict data residency within local cloud environments (Azure or AWS), without sensitive proposal data ever leaving the tenant boundary.
Because Maximus operates globally, this portability is non-negotiable. Unlike opaque SaaS AI wrappers, this architecture allows us to lift-and-shift the entire reasoning capability to where the data lives—respecting local governance while leveraging global intelligence.