Sales software can rank companies to help a team decide whom to contact first. Flight Recorder preserves the facts and rules behind a choice, so you can inspect it later.
In this example, the sales tool gives points for company facts. A total of 75 or more puts a company first in line for outreach.
Preserved factRule appliedPoints
Funding recorded on March 30 → within 90 days of the April 17 decision → adds 18 points.
Using the original rules · v3.2Illustrative replay · newer rules v5.28672points
075 · cutoff100
Other facts68Funding184
Prioritize for outreachDo not prioritize for outreach86 is above the 75-point cutoff.72 is below the 75-point cutoff.
Original recordRules v3.2 · preserved86Prioritize for outreach
Try the newer rules: the same funding fact adds 4 points instead of 18.Funding: 18 → 4 points. The four other contributions stay at 68. Newer rules also check integration pressure: Low, +0.
Choose any fact to trace its contribution. Then change the funding rule to see what happens.The facts did not change. The rules did. Flight Recorder lets you inspect the difference while keeping the original 86-point decision intact.
With the newer rules, the same funding fact adds 4 points instead of 18. The other four facts still add 68 points; integration pressure is Low and adds 0. The replay total is 72, below the cutoff of 75: do not prioritize for outreach. The original 86-point decision stays preserved. Both use the same recorded evidence.
Synthetic data. Fixed example results from the demo, not a live engine run.
Python · FastAPI · SQLite
In build
Aeverum
Your next AI tool should not have to start from zero. Aeverum carries useful memories between connected tools, with sources you can inspect and changes you control.
You are choosing a recruiter.A detail from one conversation helps in another AI tool. Then a future review suggests a connection you can judge.
A detail worth carrying.
A research assistant proposes keeping one useful detail, with the conversation attached as its source.
A new tool, without starting over.
Ask a planning assistant for help. Recall the saved detail and see how it changes the answer.
Different topics. A useful connection?
Another memory concerns a property manager. A future review could compare the incentives in both notes.
Your memory. Your verdict.
Inspect the explanation and its limits. A compelling connection still needs your judgment.
ONE MEMORY. A LONGER LIFE.MVP designFuture vision
YOU → RESEARCH ASSISTANTConversation
“This recruiter is paid when they place someone. I care about finding a hire who will stay.”
Aeverum proposes what to carry
THE DETAIL TO REMEMBERProposed
This recruiter is paid per placement.
Source: your hiring conversation. Useful when comparing recruiters.
RESEARCH ASSISTANTThe conversation you already had
SHARED MEMORYIn probation
This recruiter is paid per placement.
Source: your hiring conversation. Ready for another connected tool to recall.
PLANNING ASSISTANTA separate AI conversation
YOU ASK
“What should I ask before choosing this recruiter?”
This tool has not received your saved detail yet.
ANSWER WITH SHARED CONTEXT
“You mentioned a placement fee. Ask what happens if the hire leaves after a month.”
MEMORY 1 · HIRING
Paid when a candidate is placed.
You want a hire who stays. Source: your hiring conversation.
MEMORY 2 · HOUSING
Paid when a lease is signed.
You want a tenant who stays. Source: your housing research note.
Compare what each person is paid for with the result you actually want.
PROPOSED CONNECTION · FUTURE VISION
The fee rewards the deal. You want the outcome to last.
Different topics, a similar incentive gap. A useful question might be:what rewards a good result after the transaction?
A question to carry forward.
THE CONNECTIONBoth fees may reward a transaction, while your goal is a lasting result.
WHERE IT MAY FAILA guarantee, retention bonus, or ongoing management fee could change the incentives.
A PRACTICAL USEAsk the recruiter: “What happens if the hire leaves after a month?”
Inspect the two fictional sources
Hiring: your conversation about a placement fee.
Housing: your research note about a leasing fee.
These invented examples do not describe every recruiter or property manager.
Kept as a hypothesis for review.Its sources and limits remain attached. The connection has not become an established fact.
Connection rejected.Both original memories remain available. Only the proposed link was dismissed.
Aeverum proposes a useful detail, rather than copying the entire conversation.
Useful context travels between tools, so you do not have to explain it again.
The future vision: suggest connections across different parts of your life, with reasons you can inspect.
Keeping the idea preserves a hypothesis, not a proven fact.
MVP design: a research assistant proposes the placement-fee detail, and a second connected AI can recall it to help you ask a better question. Future vision: Aeverum could suggest a connection to your housing note, explain its limits, and let you keep it as a hypothesis or reject the link. Both original memories remain available.
Coming soon
Petasos
A trust gate for AI agents, designed to check proposed actions against fixed rules and require human approval for risky ones.
Approve a customer reply, then change one detail or lose its confirmation. See what the trust gate allows.
One changed field. A different consequence.
A useful update for Alex becomes a disclosure to the wrong person when the recipient changes. Approval must follow the action all the way to the receiver.
PETASOS / ACTION CONTROL
Interactive simulation
AI assistant→Trust gate→Supported receiver
AI assistant
Send a support reply
RecipientAlex MorganJordan Lee
Hi Alex, your replacement for order PX-1042 is on its way. Tracking will appear in your support thread when the carrier confirms collection.
→
01 / Review the consequenceReady is not permission.
Read the recipient and complete message. The assistant has proposed an action; it cannot proceed in this example until you approve it.
02 / Authority is specificOnly this action can proceed.
Approval is bound to Alex and this exact reply. The message has not been sent. Run the simulation, or test what happens when the action or confirmation changes.
Boundary test / StoppedWrong recipient. No carryover.
Jordan is now the recipient. Approval was for Alex. The changed action needs fresh review; the original approval cannot authorize it.
03 / Confirmed in this simulationApproved. Matched. Confirmed.
The simulated receiver accepted the exact approved reply and confirmed success. The record separates your approval from the execution result. No real message was sent.
Boundary test / UnresolvedNo confirmation is not a failure.
The simulated action was attempted, but confirmation was lost. The result stays unknown. Its action slot remains held, and there is no resend that could duplicate the effect.
The record behind the result
Exact approval
Waiting for you
Execution outcome
Not attempted
Action slot
Not reserved
Exact approval
Alex · reply v1
Execution outcome
Not attempted
Action slot
Ready for one attempt
Exact approval
Alex · reply v1 only
Execution outcome
Not attempted
Action slot
Fresh review required
Exact approval
Alex · reply v1
Execution outcome
succeeded
Action slot
Completed once
Exact approval
Alex · reply v1
Execution outcome
unknown
Action slot
Held · no automatic retry
Teaching record only. This is not a signed product receipt.
The earlier example is still unknown. This fresh example does not resolve it or release its held action slot.
Approve a reply for Alex, then compare three fictional paths: the exact action succeeds; a changed recipient needs fresh review; a missing delivery confirmation remains unknown and holds its action slot. A fresh example starts a separate scenario and does not resolve an unknown result.
Useful autonomy needs a boundary you can inspect: what was approved, what ran, and what is still uncertain.
Writing
Context of Humanity
Elias's publication, a briefing on power, incentives, and risk, concerning emerging tech.
Across enterprise tech, human capital management, direct sales, and operational leadership, Elias has spent his career where systems collide with human behavior.
More
Rather than staying inside a single corporate silo, his background spans mission-critical B2B platforms, direct-to-consumer financial protection, and frontline operations, giving him a grounded perspective on how organizations run and what incentives motivate across different roles and backgrounds.
Beginning in late 2023, he translated that commercial foundation directly into production code, mastering LLM orchestration and agentic engineering workflows to build the systems he always knew revenue and operations teams actually needed. By grounding narrow, advanced techniques at the frontier of applied AI within core computer science fundamentals, his architectural choices bypass the common traps of naive implementations. Having lived through the downstream fallout of broken data, disconnected tools, and fragile automations, he designs agentic workflows with strict operational discipline, knowing precisely where probabilistic reasoning unlocks leverage, where deterministic validation is non-negotiable, and how to maintain tight context economy and fault tolerance. The result is software engineered for production reality: resilient systems built to survive contact with messy data, skeptical end-users, and high-stakes operational bottlenecks.
Professional background
Scroll sideways. Click or tap a folded corner to reveal the work.
Advised mid-market leadership teams on modernizing mission-critical payroll, HR, and workforce management infrastructure at Paychex. Owned the entire commercial cycle from initial discovery to contract execution, translating operational inefficiencies, compliance liabilities, and fragmented systems into unified enterprise deployments.
Engaged enterprise design, research, and product leadership on the operational bottlenecks behind customer discovery. Translated fragmented tooling, compliance-heavy participant recruitment, and siloed research data into the foundational business case for a centralized platform, bridging the gap between messy operational workflows and executive-level software initiatives.
Built an active client portfolio from the ground up for American Income Life by turning union endorsements and client relationships into a self-sustaining referral engine. Held complete accountability for client acquisition, risk assessment, and policy placement in a 100% quota-driven environment.
Ran a high-volume bar program at Venice’s OSPI during peak service, balancing rapid drink production with attentive hospitality. Combined speed and precision with deep beverage and culinary pairing knowledge, consistently lifting check averages and guest spend while maintaining service cadence during high-demand rushes.
Promoted from the floor to run end-to-end operations and P&L for a Venice bistro and wine bar. Modernized operational workflows, renegotiated vendor contracts, and restructured staffing models to support increased nightly throughput, preserving margins, stabilizing cash flow, and steering the business through COVID-19 disruptions.
Oversaw floor execution and kitchen synchronization at Bondi Harvest, translating brand standards into daily operational rigor. Managed shift logistics, supply chain intake, and cross-team communication to deliver consistent product quality and service velocity across surge hours.
Personal Trainer & Group Fitness Instructor · 2018–2019
Designed and executed structured strength and conditioning programs at Eugene’s premier athletic facility. Built and sustained a private client roster through rigorous outcome tracking and customized programming, translating physiological goals into measurable habits that drove member retention and recurring revenue.
Ran full-cycle operations for a grassroots fundraising office partnering with major national organizations like Greenpeace and the Sierra Club. Built an end-to-end recruitment, onboarding, and coaching engine from scratch, turning zero-experience teams into high-producing frontline fundraisers capable of securing instant financial commitments on cold encounters.