Game plan

Senior Solutions Architect, Product Acceleration, AWS Worldwide Public Sector. Bill Magee's team. This page tells you where you are in the process, what the role really asks for, and exactly what to do next.

About 30 minutes to read. The box below takes 2.

Before the 5 PM call (today, 15 minutes, Zoom)

  • Goal: leave Bill thinking "this person already talks like one of my SAs," and leave yourself knowing the loop format and next step.
  • Your two sentences: "I have spent sixteen years building software in regulated industries, most recently architecting RAG and GenAI systems on Bedrock and leading a team of 15+ engineers at Prudential. Before that I spent three years at Renaissance Learning building rostering for K12 districts, so I know what an ISV selling to schools actually lives with."
  • Top 4 questions: (1) What does a great first six months look like for this hire? (2) Which customers or problems would I pick up first? (3) What is the loop format, and is there a presentation? (4) Where does Product Acceleration sit next to the account SAs?
  • The sponsorship question (ask near the end, calmly): "One logistics question so nothing surprises us later: I am on an H1B today, so I would need a transfer. Is this role open to that, and are there any US-person or GovCloud access requirements I should know about?"

The quick version

  • You are at stop one of five: a 15-minute sync with the hiring manager. Then a phone screen, the loop (about five interviews including a Bar Raiser), a debrief, and an offer.
  • The job: be the hands-on architect who helps GovTech and EdTech software companies ship GenAI features on AWS. Think "product engineer who advises," not "account manager."
  • Your strongest cards: GenAI on Bedrock in a regulated industry, AI governance and evaluation, three years in K12 EdTech at Renaissance Learning, and a year at a city agency.
  • Your real gaps: no GovTech ISV customer work yet, no public sector compliance hands-on (CJIS, FedRAMP, GovCloud). Bridge them by naming the parallel and showing you already did the homework.
  • The posting has no citizenship, clearance, or sponsorship language. Confirm with Bill anyway.
  • Every page on this site uses one running example, Beacon, a fictional GovTech and EdTech software company. Learn it once here.

Where you are

Hiring manager syncToday, 5:00 PM ET
Phone screenAbout 60 min
The loopAbout 5 interviews
DebriefWithin a week
OfferLevel and pay

Think of it like a subway line. You can see every stop from here. Each stop tests something a little different, and you only have to be ready for the next one.

StopWhoHow longWhat they testWhat good looks like
Hiring manager syncBill Magee15 minFit and interest. Can you talk to an ISV CTO? Do you understand the job?A crisp intro, one relevant story, sharp questions. He ends the call wanting to move you forward.
Phone screenUsually an SA or SA manager on the teamAbout 60 minOne or two Leadership Principle stories, plus a technical conversation (often a light design or "how would you build X").Stories with numbers and "I," not "we." A design answer that starts with questions.
The loopAbout 5 interviewers, one is a Bar RaiserAbout 55 min each, often one day on videoEach person owns two or three Leadership Principles. At least one session is a whiteboard design, and SA loops often include a customer scenario or presentation.Two strong stories per principle, a clear design walk-through, and the ability to explain it to a non-technical executive.
DebriefAll interviewers plus the hiring manager, led by the Bar RaiserYou are not thereThey share written feedback and vote.No one says "not inclined," and several say "strongly inclined."
OfferRecruiterA few callsLevel, base, sign-on, stock.You confirm level (Senior SA is usually L6), immigration support, and start date.

The Bar Raiser

The problem

A hiring manager with an open seat and a busy team is tempted to hire the "good enough" person who is available now. Over years, that slowly lowers the quality of the whole company.

Picture it

A wedding where the family is excited about the match, and one wise uncle from outside the family is asked to meet the person with no stake in the outcome. His only job is to ask, "Will this make the family stronger in five years?"

In plain words

One interviewer in your loop is from a different team. They are trained to judge whether you are better than half the people already in this role. They care most about judgment, ownership, and whether your stories hold up when they dig.

The real term

Bar Raiser. A specially trained interviewer who runs the debrief and can block a hire even if everyone else votes yes. They usually probe hard on Leadership Principles with follow-ups like "what exactly did you do," "what data did you have," and "what would you do differently."

At Beacon

If you tell a story about talking Beacon's CTO out of a risky launch, the Bar Raiser will ask: what did you actually say, what evidence did you bring, and what happened to the relationship afterwards?

Go deeper (for follow-up questions)
  • You will not always know which person is the Bar Raiser. Treat every session as if it were.
  • They listen for "I" versus "we." Say what the team did in one line, then what you did in detail.
  • They like a story that includes a mistake you owned. Have one ready (see principles.html).

The role, decoded

Each line of the posting and Bill's note, turned into plain words and what they will test.

What "Product Acceleration" means

The problem

A software company that sells to cities wants to ship an AI feature. Its engineers are good at their product but have never built with Bedrock, agents, or GenAI guardrails. The idea stalls in slides for a year.

Picture it

A visiting chef who walks into a busy restaurant kitchen, cooks the new dish side by side with the line cooks for two weeks, writes the recipe card, and leaves. The restaurant owns the dish. The chef moves to the next kitchen.

In plain words

You help a software company get from "idea" to "working prototype and a design they trust" fast. You run discovery, whiteboard the architecture, build a proof of concept with their engineers, and hand it over. They ship the production version.

The real term

AWS describes the Product Acceleration program as helping customers bring a new idea to market using modern cloud-native architectures and AI. Deliverables named in the posting: architectural proposals, deep dives, POC solutions, blogs, workshops, and driving adoption of AI-driven software development. The customers are ISVs (independent software vendors), companies whose product is software they sell to others.

At Beacon

You spend a week with Beacon's permitting team, agree on one GenAI feature (a plan-review assistant), build a Bedrock prototype in their AWS account with two of their engineers, and leave them with an architecture doc and a path to production.

Go deeper (for follow-up questions)

A sibling AWS posting for GovTech ISV SAs says plainly that the SA does discovery, foundational decisions, and prototypes but "will NOT ship production features." That is the line to respect: you accelerate, the ISV owns production. The same sibling posting lists "demonstrated use of AI-assisted development tools (Kiro, Cursor, Claude Code, or similar)" as preferred. Your Claude Code coaching practice fits that exactly.

What it saysIn plain wordsWhat they will test
"Strong experience using GenAI to develop new products and features"You have built GenAI features, not just read about them.A real build story with architecture, trade-offs, and what went wrong. A design question like "add an AI assistant to this product."
"Working directly with the leadership and engineering teams"You talk to the CTO in the morning and pair with an engineer in the afternoon.Can you explain a trade-off to an executive in two minutes, then go deep with an engineer? Earn Trust, Dive Deep.
"911 dispatch, digital evidence, permitting, benefits eligibility, K12 and Higher Ed"The customers are software companies serving governments and schools.Do you understand their world: slow procurement, strict data rules, public trust. See the next chapter.
"Architect, build, and scale modern AWS-native applications"Serverless, containers, managed databases, done well.A whiteboard design with scaling, cost, and failure handling. See design.html.
"Generative AI, agentic patterns"RAG, agents that call tools, guardrails, evaluation.When to use RAG versus an agent, how you stop an agent doing something harmful, how you measure quality. See genai.html.
"The data foundations that make both work in regulated environments"Clean, governed, access-controlled data. Where it lives and who can see it.Tenant isolation, encryption, GovCloud, CJIS, FERPA. See data-security.html.
"Show our customers how to build the next product that will delight their customers"Help them pick the right feature, not just build any feature.Customer Obsession and Invent and Simplify. Working backwards from the end user (the dispatcher, the teacher). See customer.html.
Deliverables: proposals, POCs, blogs, workshopsYou will write and teach in public, not only advise in private.Have you written, spoken, or run workshops? You have: coaching curriculum, ASCEND panels, NYU mentoring.
"Influence roadmaps" with service teamsYou bring customer pain back to AWS product teams.A story where you turned field feedback into a product change.
Basic: 8+ yrs domain, 3+ yrs design or consulting, 10+ yrs IT developmentA senior engineer who has also advised.You clear all three with room to spare.
Preferred: infra, database and networking; developer communities; exec communication; legacy migrationWell-rounded, has migrated old systems, can speak to executives.Your Vantage COLO to AWS migration and C++ to C# migration are direct evidence.
Locations: New York, Arlington, Austin. Travel 30%.New York works for you. Expect customer site visits.Say yes to travel plainly.

Who the customers are

The customer is not the city or the school. The customer is the software company that sells to the city or school. That difference matters in every answer you give.

A GovTech or EdTech ISV

The problem

A county cannot build its own 911 dispatch system, evidence locker, or permit portal. It buys one. The companies that build those products have to satisfy hundreds of cautious buyers at once, each with its own rules and budget cycle.

Picture it

A company that builds fire trucks. It does not fight fires, but every town's fire chief depends on it, every town wants a slightly different truck, and a truck that fails once makes the news.

In plain words

These are software companies, often mid-sized, that sell subscription software to governments and schools. They run many customers on one platform, they are judged on uptime and trust, and they move slower than consumer tech because their buyers do.

The real term

ISV (independent software vendor) selling multi-tenant SaaS to state and local government (SLG) and education. Compliance words you will hear: CJIS (FBI rules for criminal justice data), FedRAMP and GovRAMP (cloud security authorizations, GovRAMP was formerly StateRAMP), FERPA and COPPA (student and child data), AWS GovCloud (US) (isolated regions for regulated workloads).

At Beacon

Beacon is exactly this kind of company. See the Beacon profile below.

Real companies in this world

CompanyWhat they buildCloudSource
Tyler TechnologiesCourts, public safety, permitting, finance software for local governmentsAWS. Expanded eight-year strategic collaboration agreement, Feb 2024 (first one 2019).Tyler investor news, Feb 5 2024
Mark43Cloud-native 911 CAD, records management, analytics for police and fireAWS GovCloud. Says it is built on GovCloud, not migrated. 290+ agencies. Dallas signed Feb 2026.AWS Champions page, Business Wire Feb 2026
InstructureCanvas learning management system, K12 and higher edAWS. Built IgniteAI on Amazon Bedrock. AWS press release on Instructure, Aug 2026.AWS Public Sector blog, AWS press center
Axon (for contrast)Body cameras, Axon Evidence digital evidenceMicrosoft Azure and Azure GovernmentMicrosoft partner case study
PowerSchool (for contrast)K12 student information systemMicrosoft Azure, PowerBuddy on Azure OpenAIPowerSchool partner page

Their world and their pressures

  • Slow buyers. Cities and districts buy through RFPs and annual budgets. A sales cycle can take a year. A feature that misses a budget cycle waits another year.
  • Many masters. One product serves hundreds of agencies, each with its own rules. Data for County A must never touch County B.
  • Compliance is table stakes. CJIS for anything touching police data. FERPA and state student privacy laws for schools. Some buyers require GovCloud or a GovRAMP or FedRAMP authorization before they will even talk.
  • Thin engineering benches. Much of the team is busy keeping a large, older product running. Few have built GenAI. Hiring AI talent is hard at government-software salaries.
  • Pressure to add AI. Their customers see AI everywhere and ask for it in RFPs. Competitors announce AI features. Boards ask the CEO for an AI story.
  • Legacy code. Many products started on-premises or single-tenant. Moving to modern cloud-native is a project in itself.

Why GenAI is hard for them

WorryWhy it is sharper hereWhat you bring
Wrong answersA hallucinated benefits decision or a wrong address on a 911 call hurts real people.Grounding, human review for anything consequential, evaluation before launch.
Data leaving the boundaryEvidence and student records have legal protections. Contracts often forbid training on customer data.Bedrock does not use customer prompts to train models, models run inside the AWS boundary, GovCloud where required.
Public trustA bad AI story in the local paper can cost a contract and a council vote.Transparency, audit logs, explainable outputs, clear "AI-assisted" labels.
BudgetFixed-price contracts mean inference cost comes out of the vendor's margin.Right-size the model, cache, use retrieval where generation is not needed.
SkillsFew engineers on staff have shipped GenAI.Workshops, pairing, a reference architecture they can reuse.

Meet Beacon, the running example

Every page on this site uses the same fictional customer. Learn it once and every example will feel familiar.

FactDetail
WhoBeacon, a software company (an ISV). About 250 engineers.
GovTech customersAbout 300 US cities and counties.
GovTech products911 computer-aided dispatch, digital evidence management, a permitting portal, a benefits eligibility portal.
EdTech customersAbout 150 school districts.
EdTech productBeacon Learn, a K12 learning app.
CloudRuns on AWS. Some customers (mostly police and sheriff agencies) require GovCloud.
The askThe CEO wants to ship GenAI features within two quarters.
YouThe AWS SA assigned to help them get there safely.

What makes Beacon hard, and therefore a good practice case:

  • Two very different rulebooks in one company: CJIS for dispatch and evidence, FERPA and child privacy for Beacon Learn.
  • Multi-tenant: one platform, 450 customers, strict separation.
  • Two environments: commercial AWS for most, GovCloud for some.
  • A CEO deadline that is faster than their team's GenAI experience.

Your angle

An honest map. Lead with the strengths, name the gaps before they do, and show the bridge.

Where you are strong

What the role needsYour evidence
GenAI products in regulated settingsBuilt React and Python GenAI prototypes on AWS Bedrock at Prudential that shaped the enterprise GenAI roadmap. Built a Claude on Bedrock prototype turning sales meeting notes into CRM-ready minutes, demoed to a VP of Sales. At ITG, architected RAG pipelines and LLM integrations for regulated financial clients.
POC to production on AWSTook Amazon Kendra semantic search from proof of concept to production at Prudential, cutting 20+ steps from advisor workflows.
AWS-native architecture at scaleAdvisor Portal modernization: 8 Spring Boot microservices on AWS (EC2, S3, EventBridge, Lambda) serving 35,000+ advisors, search time from 10+ seconds to under 5.
Evaluation and responsible AIBuilt LLM evaluation harnesses (grounding, faithfulness, hallucination checks). Building PROBE, which turns AI-governance principles into testable KPIs.
Critical infrastructure AI governanceCommunity of Interest member on the NIST AI RMF Trustworthy AI in Critical Infrastructure Profile. Drafted AI incident and near-miss reporting control language that was incorporated into Discussion Draft 3 (Sep 2026).
EdTech domainThree years at Renaissance Learning (K12). Built the OneRoster rostering system on AWS that synced school districts at national scale. API performance work for K12 districts.
Government domainA year at NYC Department of Environmental Protection building city infrastructure databases.
Legacy migrationAt Vantage: on-premises COLO to Rackspace to AWS migration, and C++ to C# under live trading.
ResilienceWon the inaugural AWS Chaos Engineering GameDay hackathon at Prudential; your failure-injection standards were adopted across teams.
AI-driven development and teachingRun a Claude Code coaching practice for executives and engineers. Authored a reusable curriculum. Open-source Claude Code tools on GitHub.
Executive communicationActing CIO at Vantage. Managed 15+ engineers at Prudential. Pre-sales technical lead at ITG.
Public buildsAsk the Declaration: browser-native retrieval over the US founding documents, no server, no per-query cost. A good small story about choosing retrieval-only to remove hallucination risk.

Gaps, and how to bridge them

GapBridge in words
No GovTech ISV customer work"Financial services and public safety share the same shape: high stakes, heavy audit, customers who cannot afford a wrong answer. I have lived that shape for sixteen years. The acronyms change, CJIS instead of SEC rules, but the discipline is the same."
No hands-on CJIS, FedRAMP, GovRAMP, GovCloud"I have not deployed into GovCloud yet. I know what it is for and how it differs, and I have read the CJIS policy areas that matter for AI: access control, audit, encryption, and where data lives. I would pair with a GovCloud specialist on the first engagement." Then study data-security.html.
Not an SA title before"My consulting role at ITG is the SA job without the title: pre-sales architecture, prototypes, and teaching the client's team to own it."
Recent work is more advisory than shippingPoint at recent builds: the Bedrock CRM prototype, the RAG apps, the Ascend conference platform. Offer to walk through code.
Less recent depth in networking and databases (a preferred qualification)Use the Vantage migration and the OneRoster sync pipelines. Refresh VPC design, PrivateLink, and Aurora versus DynamoDB on design.html.
Higher ed domainRenaissance was K12. Say so plainly, then connect: "The student data rules are stricter in K12, so higher ed is a smaller step."

Your intro

Say these out loud three times each. Time yourself. The 60-second version is for Bill today. The 2-minute version is for the loop.

60 seconds

I am Swapnil Tamse, Swap. I have spent sixteen years building software where mistakes are expensive, mostly in financial services.

At Prudential I led 15+ engineers, took Amazon Kendra semantic search from proof of concept to production, and built GenAI prototypes on Bedrock that shaped the company's AI roadmap. Today I am Director of Strategic and AI Initiatives at Infinity Tech Group, where I architect RAG systems and evaluation for regulated clients, and I coach engineers and executives on AI-driven development with Claude Code.

The part that makes this role feel close to home: I spent three years at Renaissance Learning, a K12 software company, building the rostering system that synced school districts on AWS. So I have been the ISV. This job is the one I wish we had access to back then: an architect who helps a software team ship its first real GenAI feature, safely, in a regulated world.

2 minutes

I am Swap. My career has three chapters, and they all point at this role.

First, building under pressure. I spent seven years at a commodities trading firm, growing from engineer to acting CIO. I moved the company from an on-premises data center to AWS and migrated our trading systems from C++ to C# while they were live. I learned that migrations are mostly about trust, not code.

Second, the ISV side. At Renaissance Learning I built the OneRoster rostering system on AWS that synced school districts across the country, and fixed API performance for large district payloads. I saw how an EdTech company balances new features against student privacy and a very long sales cycle. Earlier, I also spent a year at New York City's Department of Environmental Protection, so I have seen government software from the inside.

Third, GenAI in regulated settings. At Prudential I managed 15+ engineers, delivered Kendra search to production, and built Bedrock prototypes for customer-facing use. My chaos engineering work won AWS's GameDay hackathon there and became a company standard. At Infinity Tech Group I architect RAG pipelines and LLM evaluation for financial clients, I am building an open framework that turns AI-governance principles into testable measures, and I contribute to the NIST AI RMF profile for AI in critical infrastructure.

What I want next is to do that work for the companies that build 911, evidence, permitting, and school software. They have the same stakes I am used to, fewer engineers to spare, and a real deadline to add AI. I would like to be the architect who helps them ship it well.

The 15-minute sync with Bill

This is not an interview in the formal sense. It is Bill deciding whether to spend his team's loop time on you. Your job is to make that an easy yes, and to learn what the loop looks like.

Goal of the call

  1. He hears a clear, relevant intro and one concrete story.
  2. You learn what success looks like and what the loop tests.
  3. You confirm the H1B transfer question without making it the headline.

The 3-part plan

MinutesPartWhat you do
0 to 4IntroThank him. The 60-second intro. Stop and let him steer.
4 to 11His questions, then yoursAnswer with one short story each. Then ask three or four questions from the list below. Take notes.
11 to 15CloseSponsorship question. Ask about next steps and timeline. One-line close: "Everything you described is the work I want to do. I would love to move forward."

Smart questions to ask (pick 4)

  1. What does a great first six months look like for the person in this seat?
  2. Which customers or problems would I likely pick up first? Is the mix more GovTech or EdTech right now?
  3. How does Product Acceleration work alongside the account SAs and the ISV partner teams? Who owns the relationship?
  4. What does a typical engagement look like end to end: how long, and what does the customer walk away with?
  5. What is the most common thing that stops an ISV from getting a GenAI feature into production?
  6. How is the team measured? Launches, adoption, something else?
  7. What does the loop look like: how many interviews, is there a presentation or a design exercise, and anything you would want me to prepare?
  8. (Near the end) "One logistics question so nothing surprises either of us later: I am on an H1B now, so I would need a transfer. Is this role open to that, and does any of the GovCloud work carry US-person requirements?"

If he asks "why AWS?"

I have built on AWS for most of my career, from our first data center migration to Kendra and Bedrock at Prudential, and I won the chaos engineering GameDay there. What pulls me now is the scale of impact. As one engineer I help one company. As an AWS architect I help the companies that serve hundreds of cities and schools, and I get to bring what I learn back to the service teams. That multiplier is what I want.

If he asks "why this team?"

Three reasons. The customers build software where a wrong answer matters, which is where I have spent my career. I have been on the ISV side at Renaissance Learning, so I know the pressure these teams are under. And the work is hands-on: discovery, design, and prototypes with their engineers, which is the part of my job I enjoy most.

"Tell me about a GenAI feature you built."

At Prudential I built a prototype on Amazon Bedrock with Claude that took unstructured sales meeting notes and turned them into CRM-ready minutes. The hard part was not the model. It was deciding what the output had to look like to be trusted by a sales VP, and making the model stick to fields we could check. I kept it narrow, one input and one structured output, so we could judge quality quickly. I demoed it live to the VP of Sales. The lesson I carry into every engagement: pick the smallest useful task, make the output checkable, and prove value before you widen it. [confirm what happened after the demo]

"What do you know about our customers?"

They are software companies that sell to cities, counties, and schools. Tyler Technologies and Mark43 on the government side, Instructure on the education side. Their buyers are slow and careful, their data is protected by rules like CJIS and FERPA, and many of them run older products while trying to modernize. Now their customers are asking for AI in RFPs, and they do not have deep GenAI benches. So they need someone who can help them pick the right first feature and build it safely. I saw a version of this from the inside at Renaissance Learning.

Questions to ask in the loop

Each interviewer usually leaves five minutes for your questions. Ask different ones in each session. Pick two per person.

For a peer Solutions Architect

  1. What was the last engagement you were proud of, and what did the customer ship afterwards?
  2. How much of your week is hands-on building versus workshops and meetings?
  3. What reusable assets does the team have, like reference architectures or sample code, and how do new ones get made?

For the hiring manager or a senior leader

  1. Where do you want this team to be in a year that it is not today?
  2. What is the hardest customer conversation the team has right now?
  3. How does field feedback from this team reach the Bedrock and agent service teams?

For the Bar Raiser

  1. What do the strongest SAs you have seen do differently in their first year?
  2. Which Leadership Principle do you see people underestimate in customer-facing roles?

For a sales or business development partner

  1. What does a great SA partner do that makes your job easier with an ISV?
  2. How do you decide which ISVs get Product Acceleration time?

Two-week prep plan

One focus per day. About 90 minutes a day. Tick things off as you go. Dates assume the phone screen or loop lands in the week of Oct 12 [confirm once scheduled].

DayFocusDo this
Mon Sep 28The syncThis page. Intro out loud 3 times. Call Bill at 5 PM. Write down what he said about the loop.
Tue Sep 29Leadership Principlesprinciples.html: read the 16 principles. Map your stories to them.
Wed Sep 30Storiesprinciples.html: write 8 STAR stories. Two per top principle: Customer Obsession, Ownership, Dive Deep, Earn Trust.
Thu Oct 1GenAI basicsgenai.html: RAG, Bedrock, Knowledge Bases. Explain each out loud with Beacon.
Fri Oct 2Agents and guardrailsgenai.html: agents, tool use, guardrails, evaluation. flashcards.html for 15 minutes.
Sat Oct 3Regulated datadata-security.html: tenant isolation, encryption, GovCloud, CJIS, FERPA.
Sun Oct 4Rest and reviewFlashcards only. Say the 2-minute intro once.
Mon Oct 5System design methoddesign.html: the whiteboard method. Practice one Beacon design out loud, 35 minutes, timed.
Tue Oct 6Design practicedesign.html: a second design (Beacon Learn tutor or permit assistant). Draw it by hand.
Wed Oct 7Customer skillscustomer.html: discovery questions, pushing back, working backwards.
Thu Oct 8Mock loop, part 1Two 45-minute mocks: one LP, one design. Record yourself. Watch it back.
Fri Oct 9Fix weak spotsRedo the weakest story and weakest design. flashcards.html full deck.
Sat Oct 10Presentation (if the loop has one)Build a 10-minute talk: "How Beacon ships its first GenAI feature in 90 days."
Sun Oct 11Light dayIntro once. Questions list once. Sleep early.

Day-of checklist

Setup (the night before)

  • Zoom updated, camera at eye level, light in front of you, plain background.
  • Wired internet if possible. Phone hotspot ready as backup.
  • Notifications off on laptop and phone.
  • Water and a snack within reach.

Whiteboard tool

  • Ask the recruiter which tool they use (often a shared online whiteboard). Practice drawing boxes and arrows in it once.
  • Backup: paper and a thick marker, held up to the camera.
  • Start every design by writing the requirements in the top corner.

Notes on the desk

  • One index card: your 8 stories by title and the principle each covers.
  • One index card: your design opening questions (users, scale, data rules, latency, cost).
  • The questions list for each interviewer.
  • Do not read from notes while answering. Glance, then look at the camera.

Energy

  • A real meal before the loop. Protein, not just coffee.
  • A short walk before the first session.
  • Between sessions: stand up, water, one deep breath. Do not replay the last session.
  • If one session goes badly, the next interviewer does not know. Reset.