Reasoning methods
Produces a memo and an externally checkable claim and source packet.
Vassiliy
Lakhonin
I turn source packs into reviewable documents, and build systems that record what is sourced, what is missing, and who must review it.
Almaty, Kazakhstan · UTC+5
Program and portfolio operations since 2014
Two source records
[1] Project status Signed; loan up to €35m.
[3] Later notice: US$20m loan plus US$5m internal cash for the same Project ID.
- Confirmed
- Loan record and stated project status.
- Conflict
- The financing figures cannot yet be reconciled.
- Human check
- Obtain the governing facility, grant, and supersession records before reliance.
Public-source worked sample, not client work. Not a factuality verifier or financing approval. No autonomous live source retrieval. Not legal, compliance, sanctions, financial, investment, insurance, or trading advice. Human review is required before commercial action.
Two ways I work
The operated work produces a document. The open-source work makes the review logic inspectable.
-
Documents prepared from your files
Corridor and infrastructure documents, Russia or CIS trade-file reviews, and grant application packages. The scope, source gaps, correction cycle, and human owner stay visible.
-
Review infrastructure you can inspect
Agenda Intelligence MD checks caller-supplied claim and source packets before human review. Three specialist repos define the reasoning methods that feed the same packet contract.
Proof, with its limits attached
Public artifacts are separated from self-reported delivery records and unvalidated workflow hypotheses.
-
Aktau bankability screen
A worked document built from five public primary records. It shows claim-to-source traceability and written gaps.
Illustrative sample. Workflow awaiting external validation.
-
Agenda Intelligence MD
Open-source claim and source packet checks, versioned contracts, local validators, and CI.
Public source. Packet completeness, not factual truth.
-
Donor reporting
Multi-country reporting QA, partner capacity, and evidence-pack delivery from prior program work.
Self-reported. Evidence path available in the case record.
-
GrantFlow
Bid record, donor-aware draft, logframe, MEL plan, reviewer queue, evidence gaps, and exports.
Public source. No paying case recorded.
One packet, three review layers
The public stack keeps reasoning, structural checks, and the final human decision separate.
Packet check
Agenda Intelligence MD checks references, declared quotes, lexical support, and unmatched numbers.
Human review
The named reviewer decides what proceeds, what is held, and what needs another source.
No autonomous approval.
Program delivery. AI systems.
I have worked across donor-funded program delivery, monitoring and evaluation, audit readiness, and cross-border portfolio operations since 2014. My most recent employed role was Regional Monitoring and Evaluation (Performance) Manager at DAI on a USAID-funded multi-country portfolio from 2023 to 2025.
Since February 2026 I have built Agenda Intelligence MD and the workflows shown here full-time. The fit is program operations, evaluation operations, agent workflow implementation, evidence QA, and strategic-risk research support. I do not position as an ML research scientist.