AI development partner requestOriginal source · buyer request
RESEARCHING✓ Buyer✓ Demand✓ Source✓ Fit
DOTENUM INTERNAL AI SYSTEMActive internal development
GrowthOS
Evidence-grounded prospect intelligence powered by multi-stage AI research.
An internal system built by Dotenum to investigate potential business opportunities across noisy public information—combining AI reasoning, source evidence, deterministic controls, and human review.
Sanitized interface reconstruction
G
Demo dataGrowthOSInternal intelligence system
Workspace
⌂Overview
⌕Discovery
◇Prospects
▦Pipeline
◎Agents
Evidence-driven discovery
3 candidatesQualification queue
AI services procurement noticeAuthorized source · deadline passed
DISQUALIFIED✓ Buyer✓ Demand✓ Fit× Expired
AI startup directorySecondary source · no buyer intent
DISQUALIFIED✓ Editorial× No demand
Discover›
Investigate›
Verify›
Qualify›
Review›
Pipeline
Multi-stage AI researchEvidence groundingDeterministic gatesHuman-in-the-loop
01 / THE PROBLEM
Search results aren’t prospects.
Search engines return articles, vendor directories, supplier advertisements, old procurement notices, vacancies, funding news, and companies merely discussing AI.
GrowthOS starts with a stricter assumption: a candidate is not a prospect until the evidence supports it.
01×
AI vendor listicle
Editorial
02×
“We build AI applications”
Supplier advertisement
03×
AI funding announcement
No buyer demand
04×
AI engineer vacancy
Employment request
05×
AI services RFP
Deadline passed
06✓
External AI partner requested
Potential buyer request
Search resultNot yet a prospect
02 / RESEARCH SYSTEM
Research first. Qualify second. Automate carefully.
GrowthOS decomposes prospect intelligence into independent stages. Each stage answers a different question and preserves the evidence behind its conclusion.
01User intent›
02Research planner›
03Discovery engine›
04Candidate extraction›
05Source investigation›
06Buyer & role analysis›
07Demand extraction›
08Capability matching›
09Qualification gates›
10Human review›
11Prospect pipeline
Evidence storeSource provenanceAudit trailDuplicate protectionContact verification
RESEARCH PLANNER
One request becomes a research strategy.
GrowthOS generates targeted research themes from the user’s intent, then records the query strategy, source, market, date window, yield, and stopping reason.
User intent
“Find companies looking for external AI development help.”
AI development partnerAI implementation vendorAI automation partnerRAG developmentAI agent implementationDocument intelligenceBackend/API integrationSoftware subcontracting
03 / EVIDENCE
Every important conclusion should trace back to evidence.
Discovery evidence and qualification evidence are intentionally different. A snippet may reveal something worth investigating; it does not automatically prove the underlying claim.
EVIDENCE RECORDSUPPORTED
- Claim
- External AI development services requested
- Evidence excerpt
- “Seeking a qualified delivery partner to design and implement an AI-enabled operational workflow…”
- Source
- Original / authorized source
- Provenance
- Preserved with research record
01Original sourceQualification evidence
02Authorized source / exportQualification evidence
03Secondary sourceCorroboration
04Search snippetDiscovery only
Model-generated summaries remain separate from the excerpts and provenance used to support a decision.
Knowing who said what matters.
Participants are modelled by their role in the specific opportunity: buyer, representative, supplier, intermediary, publisher, mentioned entity, or unknown.
Post authorPotential buyer
Company ABuyer organization
Company BResponding AI supplier
Unknown remains unknown.
Instead of “this looks promising,” GrowthOS represents the request as structured facts and retains missing commercial information as missing.
- Requested deliverable
- Custom AI application
- Buyer problem
- Operational workflow automation
- Engagement
- External vendor requested
- Timeline
- Known
- Budget
- UNKNOWN
- Submission method
- Known
06 / QUALIFICATION
AI reasoning constrained by deterministic gates.
A candidate is admitted only when all eight mandatory requirements pass. The rules preserve business constraints that a plausible model answer cannot override.
01
Identifiable buyer
Can the requesting organization or person actually be identified?
02
Explicit external demand
Is there evidence they want an outside company, vendor, or team?
03
Source integrity
Can the underlying request be verified from an adequate source?
04
Current & actionable
Is the opportunity still open, current, and possible to act on?
05
Service fit
Does the requested work match Dotenum’s capability registry?
06
Engagement compatibility
Is the permitted engagement compatible with a delivery company?
07
Legitimate contact route
Is an authorized response route tied to the buyer and this request?
08
Safety & duplicate check
Is there a disqualifier, prior contact, or duplicate opportunity?
PASSFAILUNKNOWN
A confidence score cannot override a failed mandatory gate.
07 / ACTIONABILITY
Strong fit doesn’t override failed actionability.
A buyer can be real. The demand can be explicit. The work can match Dotenum perfectly. If the verified deadline has passed, it is not an actionable opportunity.
Qualification result · demo dataDISQUALIFIED
BuyerVerified
✓External demandVerified
✓AI service fitStrong
✓SourceEvidence available
✓ActionabilityDeadline passed
×ContactUnknown
?Final decision
DisqualifiedReason: opportunity expired
08 / TRY THE LOGIC
Three signals. Three different outcomes.
This lightweight simulation is disconnected from the internal GrowthOS application. It shows how the same qualification model handles a supplier advertisement, an expired buyer request, and an active case with unresolved contact evidence.
G
Qualification reviewGrowthOSInteractive simulation · demo data
Choose a research candidate
Qualification decision
DISQUALIFIEDCustom AI services request
Sanitized representative case · no live system connection
G1 · BuyerPASS
G2 · DemandPASS
G3 · SourcePASS
G4 · ActionableFAIL
G5 · Service fitPASS
G6 · EngagementPASS
G7 · ContactUNKNOWN
G8 · SafetyPASS
Primary reasonOpportunity deadline passed
Evidence summary
Buyer, external demand, source and capability fit are supported. The authorized source confirms that submissions have closed.
A failed mandatory gate determines the result even when other evidence is strong.
View audit trail09 / CONTROLLED AUTOMATION
Automation where it’s safe. Human judgment where it matters.
Ambiguous evidence is escalated rather than converted into a confident hallucination. Research and outreach remain separate responsibilities, and outbound action stays behind explicit controls.
Resolve the ambiguity, not just the status.
Request more researchCorrect buyerCorrect roleCorrect evidenceRequalifyRejectSuppress
See how the research happened.
Discovery✓ Completed
Candidate extraction✓ Completed
Source verification✓ Completed
Buyer resolution✓ Completed
Demand analysis✓ Completed
Capability matching✓ Completed
Qualification✓ Completed
Research and outreach remain separate.
1Qualified
2Contact verified
3Duplicate check
4Human approval
5Outreach ready
Sending is disabled by default in the current internal system and remains subject to approval, opt-outs, and limits.
01Search resultsResearch input›
02Research candidatesNot prospects›
03QualifiedAll gates pass›
04Outreach readyHuman-controlled
10 / IMPLEMENTATION
A product surface backed by an auditable research system.
The public architecture below reflects the current repository and local deployment—not a generic AI stack.
01 · InterfaceReact application↓
02 · ApplicationFastAPI + SQLAlchemy APIs↓
03 · OrchestrationCelery research and scheduled workers↓
04 · IntelligenceAzure OpenAI + structured extraction↓
05 · DiscoveryLocal SearXNG public-source search↓
06 · StatePostgreSQL + Redis↓
07 · ControlsGate engine + human approvals
ReactTypeScriptFastAPISQLAlchemyPostgreSQLRedisCelerySearXNGAzure OpenAISMTP / IMAP
Source provenance
Audit history
Capability registry
Deduplication
Contact association
Research state
11 / ENGINEERING CHALLENGES
The difficult work sits between a signal and a safe decision.
Search relevance isn’t buyer intent.
Intent-oriented discovery followed by source-backed verification.
The company mentioned may not be the buyer.
Opportunity-specific participant and role modelling.
Missing facts can become plausible inventions.
Structured fields that preserve an explicit UNKNOWN state.
A legitimate project can still be unusable.
Deadline and current-actionability verification.
A contact can belong to the wrong company.
Contact-to-buyer and contact-to-request association checks.
AI confidence can hide hard failures.
Mandatory deterministic gates that confidence cannot bypass.
Automation can create duplicate or unsafe outreach.
Deduplication, suppression, approval, and outbound safety controls.
12 / BEYOND SALES
GrowthOS is one implementation. The architecture applies much further.
The reusable capability is not a lead generator. It is the pattern: research, structure, verify, apply business rules, retain evidence, and route uncertainty to people.
Procurement intelligence
Find and qualify relevant tenders.
Vendor intelligence
Research suppliers against business requirements.
Compliance research
Collect evidence and escalate uncertain cases.
Document operations
Extract, validate, and route information.
Customer operations
Use agents while preserving business controls.
Knowledge intelligence
Research across approved internal and external sources.
DEVELOPMENT TRANSPARENCY
Status: active internal development
GrowthOS is presented as an internal engineering case study—not a commercially available SaaS product or a benchmark-proven replacement for human prospecting.
No performance percentages, customer logos, fabricated revenue, or autonomous-outreach claims are used. Evaluation metrics will only be published after an authorized benchmark.
BUILD WITH DOTENUM
Have a workflow that needs more than a chatbot?
Dotenum designs custom AI systems that combine agents, business logic, integrations, structured data, evidence, and human oversight.