RFQ Document Inference
Extract data from unpredictable data dumps by populating a benchmark ontology which updates in real time in light of new information.
RFQs are Data Intensive
Every customer has a unique data representation which resists repeatable RFQ automation agnostic to differing conventions and jargon
Financial Modeling for RFQs require understanding relevant sections of bulky data
Every project has its own representation of relevant data coming from customization on the company's ground truth
Even this ground truth has drastic variation across companies
Fundamentally, this data and the manipulation thereof is informed by the same engineering principles but there is neither knowledge eagerness to adapt a universal ontology for RFQ data across organizations
Current Bottleneck
Orphic must build custom parsers and interpreters for each customer providing a unique represented data with varying focus and relevance
Atlas Solution
Atlas maintains a living universal schema for all slices of interest across spreadsheets, PDFs, and JSONs, identifying and extracting relevant fields with confidence.
Security & Data Governance
This operates on synthetic data. Atlas does not have access to any additional data.
Pilot guardrails (default)
- Internal data from companies will not be compromised
- No training on Enterprise data; configurable retention/deletion
- Role-based access; audit logs for every access and export
We can tailor controls to Orphic's internal AI pilot review process; the pilot does not require deep integration.
Pilot Overview
4 weeks to understand 6 Data Dumps
Each data dump is its own project. The goal is to know what is relevant in each project.
RFQ Data Dump
Extraction Memo
API Integration
Orphic provides 6 data dumps with different RFQ use cases containing Spreadsheets, PDFs, and JSONs
Atlas converts each file into Natural Partitions with its corresponding extractions
The identified partitions and their schematized extractions are available via API integration
Schemas for Natural Partitions can enhance over time with the complete extraction history available
Impact Tracked
Primary: Schema Formation with extraction correctness
Ontology
Defined structure for speciic types of information
e.g. Knowing what a 'Grade-based-Price for X' is and identifying the relevant aspects for it across all occurrences
Baseline
Non-Existent
Target
Stable inferred schema
Extraction Confidence
Likelihood of correctness for all extractions
Review Count is the number of observations which need to be reviewed to identify and correct all mistakes
Baseline
Non-Existent
Target
95% Accuracy, 10% Review Count
Week-by-Week Structure
3 weeks to decision
Setup + Definitions
- Atlas provides all Data Dumps
- Agree on pilot guardrails and data handling
- Define export format for case packets (ticket + attachments + relevant exports)
Output: Signed-off pilot plan + data schema
Baseline + First Run
- Run Atlas on first data dump
- Feedback session for necessary adjustments in output
- Mid-pilot check-in
Output: First-run metrics + Understanding of insight gap
Iteration + Second Run
- Incorporate reviewer feedback (what was missing / misleading)
- Feedback on Insights from first 3 Data Dumps
- Finalzie scope and variation for post-pilot adoption
Output: Delta metrics + updated artifacts
Results + Go/No-Go
- Compile report, identify best-fit teams, define next scope if positive
- Additional Data Dumps Processed
Output: Go / no-go decision + rollout plan
Engagement Includes
4-week pilot investment
Includes
- Pilot setup + investigation packet schema alignment
- Versioned schema memo + must-verify evidence set generation
- Change log evaluation (investigation update injection or second batch)
- Security/retention configuration + LLM provider options
- Weekly review + iteration loop with investigation team
- Final report + go/no-go recommendation + auditability assessment