Playbooks Career navigation Role transition roadmaps
SWE → AI Engineer Resume impact language Salary negotiation scripts Personal brand for AI
Role transition roadmaps

SWE → AI Engineer

8 min read · Updated March 2026 · Free playbook
Power tip
Your coding skills are the moat — do not apologize for them. Every AI Engineer who cannot write clean async Python, design a reliable API, or think in system components is a liability in production. You already have the hardest part. The gap is domain knowledge, not engineering ability.

Software engineers are the strongest candidates for AI Engineering roles — not despite their background, but because of it. AI systems fail in production for the same reasons software systems fail: poor error handling, no monitoring, untested edge cases, and no rollback strategy. Engineers who understand these failure modes build more reliable AI than researchers who don't.

The gap from SWE to AI Engineer is narrower than most people think. You need to add three things: working knowledge of how LLMs behave and fail, hands-on experience with the core AI engineering patterns (RAG, agents, evals), and one shipped project that demonstrates you can apply them. That's a focused 60–90 day build, not a career restart.

The 90-day plan

Days 1–30Build the foundation
  • Complete fast.ai Practical Deep Learning Part 1 — gives you intuition for how models learn without requiring PhD math
  • Read the Anthropic and OpenAI documentation end to end — not tutorials, the actual API docs. Understand tool use, context windows, and streaming.
  • Build a simple RAG pipeline from scratch: ingest a PDF, chunk it, embed it, store in a vector database, retrieve on query. Use LlamaIndex or LangChain.
  • Learn the basics of prompt engineering — chain-of-thought, few-shot, system prompts, output constraints. Andrej Karpathy's intro video is the best 2-hour investment.
  • Set up LangSmith or Braintrust for eval tracking — even on a toy project. Eval infrastructure from day one is a senior signal.
Days 31–60Build the portfolio project
  • Pick one real problem from your SWE background and build an AI solution for it. The domain context is your advantage — use it.
  • Implement a proper eval suite before you build the feature. Define what "good output" means, build a test set, measure before you optimize.
  • Add observability: log every LLM call, track latency, measure token costs. Production AI engineering is 40% infrastructure.
  • Implement at least one non-happy-path scenario: what happens when the LLM returns malformed output? When the vector DB has no results? When the API rate limits?
  • Write a README that explains your architecture decisions — why RAG vs fine-tuning, why this chunking strategy, what the eval results showed.
Days 61–90Position and apply
  • Publish the project publicly. Write a LinkedIn post explaining what you built, one thing that surprised you, and one thing you'd do differently.
  • Update your resume to lead with AI system outcomes — not "used LangChain" but "built RAG pipeline that reduced hallucination rate by 40% on domain-specific queries vs baseline."
  • Study AI system design interview formats — the most common prompt is "build a document Q&A system for [company]." Practice narrating your architecture decisions out loud.
  • Apply to AI Engineer roles at companies where your domain background is an advantage. A fintech SWE applying to a fintech AI company has a story a generalist doesn't.

Skills to add — what matters vs what doesn't

Must add
LLM API integration
OpenAI, Anthropic, Google — prompt construction, context management, tool calling, streaming. The foundation of every AI engineering role.
Must add
RAG architecture
Chunking, embedding, retrieval strategies, reranking, eval. The most common AI engineering interview topic in 2026.
Must add
Evaluation frameworks
RAGAS, LLM-as-judge, human eval pipelines. AI Engineers who can't measure quality can't improve quality. Evals are the job.
Must add
Vector databases
Pinecone, Weaviate, pgvector. Understand the tradeoffs — interviewers ask this specifically.

Portfolio projects that signal readiness

Project 1 — RAG pipeline with production-grade evals
Build a retrieval-augmented generation system over a real document corpus (company docs, legal filings, medical literature). The differentiator: include a comprehensive eval suite with RAGAS metrics, human eval samples, and retrieval precision/recall tracking. Deploy with LangSmith or Braintrust observability. This project signals: you understand the most common AI engineering pattern, you can evaluate quality rigorously, and you think about production monitoring from day one.
The eval suite is what separates this from a tutorial project. Every junior candidate builds RAG. Almost none measure whether it actually works well.
Project 2 — AI agent with tool use and error handling
Build an agent that uses multiple tools (web search, database queries, API calls) to accomplish a multi-step task. Focus on: graceful failure when tools return errors, retry logic, cost tracking per request, and a clear system prompt architecture. This project signals: you can build complex AI systems that handle real-world messiness, not just happy-path demos. Tool-use agents are the fastest-growing category of AI engineering work in 2026.
Include a cost-per-request dashboard. It signals production thinking and is the first question any engineering manager will ask about an agent system.
Project 3 — LLM-powered internal tool (domain-specific)
Build an AI tool that solves a real problem from your current or previous SWE role. Examples: automated code review assistant, incident response summarizer, customer support ticket classifier. Use your domain context as the differentiator. This project signals: you can identify where AI adds genuine value in a software workflow, scope the problem correctly, and build something colleagues would actually use.
Domain-specific projects are 3x more compelling in interviews than generic chatbots. Your SWE background IS the advantage — use it.
Project 4 — Open-source contribution to an AI framework
Contribute to LangChain, LlamaIndex, Instructor, or another AI engineering framework. Target: documentation improvements, bug fixes, or small feature additions. The bar is lower than you think — many AI frameworks have open "good first issue" tags. This project signals: you can work in a professional codebase, collaborate through PRs, and understand the tools the industry uses. It also gives you something concrete to discuss when interviewers ask about your engagement with the AI ecosystem.
Even a documentation PR counts. The signal is community engagement and professional engineering practices, not the size of the contribution.

How to position your SWE background in interviews

When they ask "why are you moving from SWE to AI Engineering?"
Lead with the production problem, not the technology excitement. "I kept running into the limits of deterministic systems when dealing with unstructured data at [company]. RAG and LLM-based approaches solve a class of problems that traditional software can't — and my engineering background means I can build them reliably, not just demo them."
Excitement about AI is everywhere. Production credibility is rare. Lead with reliability, not enthusiasm.
When they ask "do you have ML experience?"
"I've been working with LLM APIs and building RAG systems for 3 months — I can show you the project and the eval results. I don't have model training experience and I'm not positioning for that. What I bring is production engineering rigor applied to AI systems, which is where most AI products actually fail."
Honesty + specificity + a clear value proposition beats vague claims every time. Knowing what you're not positioning for is a sign of self-awareness.
When they give you a system design prompt
Start with requirements before touching architecture. "Before I design the system, I'd want to clarify: what's the latency budget? Is the data corpus static or changing? What's the acceptable hallucination rate? What does a failure look like to the user?"
This set of questions signals production thinking. Candidates who jump straight to "I'd use LangChain and Pinecone" signal demo thinking. The questions are more impressive than the answer.
🎯
Practice makes permanent — InterviewFlo
Now practice delivering these answers with real AI Engineer interview questions
InterviewFlo generates personalized questions matched to your resume and target company — with instant AI feedback on your actual answers.
Resume by role

Impact language guide

10 min read · Updated March 2026 · Paid playbook
Power tip
Recruiters spend 6 seconds on a resume before deciding whether to keep reading. In those 6 seconds they scan for three things: the role title, the company name, and one number. If there is no number in the first bullet of each role, there is no hook. Every bullet that starts with "Responsible for" is wasted real estate.

The single highest-leverage resume change is converting task descriptions into outcome statements. A task tells them what you did. An outcome tells them why it mattered. The formula: [Action verb] + [what you did] + [measurable result] + [business context]. Not every bullet needs a metric — but every bullet needs a "so what."

Before and after — by role type

Data Analyst
Before
Responsible for building dashboards for the sales team
After
Built 14 executive dashboards used by 120+ sales reps, reducing weekly reporting prep time from 4 hours to 20 minutes
Adds scope (14 dashboards, 120 users), quantifies impact (time saved), and makes business value obvious.
Before
Analyzed customer churn data and presented findings to leadership
After
Identified 3 early churn signals from behavioral data analysis, enabling a proactive outreach program that recovered $340K in ARR in Q3
Connects the analysis to a business outcome (revenue recovered) and shows the causal chain between your work and the result.
AI / ML Engineer
Before
Built a machine learning model to predict customer churn
After
Developed XGBoost churn model (AUC 0.89) deployed to production, enabling proactive outreach that reduced monthly churn by 12% and saved $840K annually
Includes the technical credential (AUC) for technical readers AND the business outcome for business readers. Both audiences are in the room.
Before
Worked on LLM-based document processing pipeline
After
Built RAG pipeline processing 50K documents/day at 94% retrieval precision, replacing a manual review process that cost $180K/year in contractor time
Shows scale (50K docs/day), includes a quality metric, and quantifies the business case (cost replaced).
AI Product Manager
Before
Led development of AI-powered search feature
After
Defined eval framework for AI search redesign; shipped to 2M users with 31% improvement in task completion rate and 18% reduction in zero-result searches
Leads with the PM-specific contribution (eval framework) before feature details. Shows both scope and measurable user outcome.
Data Engineer
Before
Built ETL pipelines to ingest data from multiple sources
After
Architected ingestion pipelines processing 4TB/day from 12 source systems with 99.8% uptime SLA, enabling real-time analytics for 200+ internal users
Quantifies scale (4TB, 12 sources), includes reliability metric, and shows downstream value (real-time analytics).
AI Consultant / No-code AI
Before
Helped organizations implement AI tools
After
Led AI adoption program for 1,200-person retail organization, achieving 78% active user rate within 90 days — 3x the industry average — through a champion network of 24 internal advocates
Shows scale (1,200 people), quantifies adoption rate, benchmarks against industry, and credits the method that drove the result.

Action verb bank

Built / created
ArchitectedBuiltDesignedDevelopedEngineeredLaunchedShipped
Led / managed
DirectedLedManagedOrchestratedSpearheadedSteered
Improved / optimized
AcceleratedImprovedOptimizedReducedStreamlinedTransformed
Analyzed / surfaced
AnalyzedDiagnosedEvaluatedIdentifiedInvestigatedSurfaced

When you don't have a metric

Situation
You don't know the exact number
Solution
Estimate with a qualifier: "reduced reporting time by approximately 3 hours per week." Approximate numbers are better than no numbers.
Situation
The impact was qualitative
Solution
Use scope as a proxy: "Used by 200+ analysts across 8 business units." Adoption is evidence of value.
Situation
The project is ongoing
Solution
Use leading indicators: "pilot phase showed 34% improvement in Z, rolling out to full user base." Progress is still impact.
Situation
The work was foundational infrastructure
Solution
Describe what it enabled: "Built data pipeline that enabled real-time dashboards for the first time across 3 product lines." The downstream capability is the impact.
Salary & comp benchmarks

Salary negotiation scripts

12 min read · Updated March 2026 · Paid playbook
Power tip
"I'm really excited about this role — based on my research and the scope of what we've discussed, I was expecting something in the range of X to Y. Is there flexibility there?" This single sentence outperforms every other negotiation opener. It signals enthusiasm, anchors with research, gives a range not a number, and ends with an open question. Practice it until it sounds natural.

Negotiating a salary offer is one of the highest-return activities in your career. A 10% improvement on a $150K offer is $15,000 in year one — compounding through every future raise and offer benchmark. Most people leave this on the table not because they lack leverage but because they don't know what words to say.

These scripts are frameworks to internalize, not templates to read verbatim. Read each one several times, then practice saying it out loud until it feels like your own language.

The 6 negotiation scripts

Script 1 — Responding to the first offer (phone)
Use when a recruiter calls to deliver the offer verbally. Never negotiate in the moment — always buy time first.
"Thank you so much — I'm genuinely excited about this opportunity and the team. I want to give this the consideration it deserves before responding. Can I have until [48–72 hours from now] to review everything and get back to you?" [They confirm. Then 24–48 hours later, call back:] "I've had a chance to look at everything carefully. I'm really enthusiastic about joining — the role is exactly what I'm looking for. I do want to have a transparent conversation about compensation. Based on my research into market rates for this role at this level and company type, I was expecting something in the [your range] range. Is there flexibility to get there?"
Always ask for time. The pause gives you time to research and gives them time to feel the offer might not land — which creates pull before you even negotiate.
Script 2 — Email counter after reviewing the written offer
Use for the written negotiation — cleaner than phone for Data & AI roles because it gives them time to discuss internally without pressure.
Subject: Re: Offer — [Your Name] Hi [Name], Thank you for the formal offer — I've reviewed everything carefully and I'm genuinely excited about the role and what the team is building. I want to be transparent: I'd like to discuss the base salary. Based on my research into current market rates for [role] at this level — Levels.fyi, recent offer data from similar companies, and community sources — I'm seeing base compensation in the [range] range for comparable roles. I'm confident in the value I'd bring, particularly [one specific reason]. Would you be able to get to [specific number] on base? I'm committed to making this work and look forward to your thoughts. [Name]
Name your data sources (Levels.fyi, market research). It signals preparation. Include one specific value statement — not a list of reasons, one strong one.
Script 3 — You have a competing offer
Use when you have a real competing offer. Never fabricate one — it backfires when they ask for details.
"I want to be transparent with you because I'm genuinely most excited about this role. I've received another offer from [company type] for [amount]. I'm not using it as a tactic — I'd prefer to be here. But there's a gap I need to close to make the math work. Is there flexibility to get to [target]? I'd be ready to sign today if we can get there."
"I'm not using it as a tactic" paradoxically lands better — it sounds honest. "I'd be ready to sign today" creates urgency without being pushy.
Script 4 — Negotiating equity (startup or growth stage)
Use when the base is firm but equity is flexible — common at Series A-C companies.
"I understand the base is where it is and I'm not trying to push on that. What I'd like to discuss is the equity component. Given [the vesting cliff / the current valuation / the risk profile], I was hoping for something closer to [your target range]. Is there flexibility on the equity package, or on the vesting schedule?"
Many companies have more flexibility on equity than base. Mentioning the vesting schedule opens a second dimension — even if they can't change the grant size, they might offer accelerated vesting.
Script 5 — When they say the offer is firm
Use when the recruiter says "this is our best and final offer."
"I appreciate the transparency. I want to make this work — I'm excited about the role and the team. If the base is truly fixed, can we talk about one alternative: a signing bonus to bridge the gap, an earlier performance review at 6 months with a defined path to [target base], or an additional week of PTO? I'm flexible on structure if there's a way to close the gap."
"Best and final" is often a negotiating position, not a fact. A signing bonus doesn't affect salary bands and is the most common alternative currency. Pivoting to alternatives demonstrates flexibility and often unlocks movement.
Script 6 — Counter-offer from current employer
Use when your current employer makes a counter-offer after you resign.
"I genuinely appreciate this — and I want to be honest with you about why I'm leaving. This isn't primarily about compensation. It's about [the real reason — growth, technology, scope]. A counter-offer addresses the symptom. If the underlying things that matter to me change, I'm absolutely open to that conversation. But I wouldn't want to accept more money and leave in 6 months anyway — that wouldn't be fair to either of us."
Accepting a counter-offer keeps you an average of 6 months before you leave anyway — the reasons you wanted to leave rarely change. This script lets you decline gracefully without burning the relationship.

The 5 mistakes that cost the most

Mistake
Giving a number before they do
Why it costs you
You anchor too low or too high. Always try to get their range first. If they press, give a range with your target at the bottom.
Mistake
Negotiating against yourself ("I know it's probably not possible but...")
Why it costs you
Removes your leverage before you start. State the ask directly and let silence do the work after.
Mistake
Accepting or rejecting verbally in the moment
Why it costs you
Decisions made under social pressure are worse decisions. Always ask for 24–48 hours. It's expected and respected.
Mistake
Treating negotiation as adversarial
Why it costs you
The framing that converts: "I want to make this work — here's what I need to get there." You are solving a problem together.
Mistake
Stopping after one round
Why it costs you
Most offers move twice before they're genuinely final. Two counters is where most people stop. Three is where the real money is for senior roles.

The most underused tool: silence

After you make your ask — stop talking
Most people fill the silence immediately with qualifications, apologies, or backtracking. The silence feels unbearable. Resist it. The person who speaks first after a salary ask is almost always the one who concedes. Say your number. Then wait. Let them respond. Whatever they say next is information — and it will usually tell you whether there's room to move before they tell you explicitly. The pause that feels like 30 seconds to you feels like 5 seconds to them. It is not awkward. It is professional.
Personal brand

Personal Brand for AI

12 min read · Updated April 2026 · Free playbook
Power tip
The highest-performing AI content format is "Here's what I learned failing at X." Practical failure posts consistently outperform tutorials in reach and inbound. Start sharing before you feel ready — audiences compound around the journey, not the destination.
Quick win — do this in the next 30 minutes
Write your first LinkedIn post using this template: "I tried [X]. I expected [Y]. What actually happened was [Z]. Here's what it means for [audience]." Schedule it for tomorrow morning. That's it. The first post is the hardest — this format removes all the friction.

Your personal brand is not a vanity project — it's a compounding career asset. In AI, where hiring managers can't easily verify skills from a resume alone, your public body of work becomes your strongest signal. A well-positioned LinkedIn presence, a handful of technical posts, and one conference talk will generate more inbound opportunities than 200 cold applications.

The goal isn't to become an influencer. It's to make yourself findable and credible when someone searches for expertise in your area. Every piece of content you publish is a permanent node in the network that connects you to opportunities.

1. LinkedIn content strategy

The 5 post formats that perform in AI

Not all content is created equal. The formats that consistently generate reach and inbound in AI are: failure stories ("I tried X, here's what went wrong"), framework shares ("The 3-step process I use for Y"), hot takes ("Unpopular opinion: Z is overrated"), behind the scenes ("Here's what building an AI product actually looks like day to day"), and results breakdowns ("We shipped X, here are the real numbers"). Stick to these five formats and rotate through them. You'll never run out of ideas.

Hook writing — the first line determines everything

LinkedIn truncates posts after the first 2 lines. If your hook doesn't create a reason to click "see more," the rest of your post doesn't exist. The best hooks create a knowledge gap: "I spent 3 months building RAG pipelines. Here's the one mistake that cost me the most time." or "Everyone's talking about AI agents. Almost nobody is talking about the reliability problem." Write the hook last — after you know what the post actually says — then make it the most surprising or specific thing in the piece.

Posting cadence: biweekly beats daily burnout

The biggest mistake new content creators make is posting every day for two weeks and then disappearing for three months. Consistency beats frequency. Two well-crafted posts per week — every week — will outperform daily posting that fizzles after a month. Tuesday and Thursday mornings perform best for AI/tech audiences. Batch-write on weekends, schedule during the week. Protect the cadence above everything else.

Why failure posts outperform tutorials

Tutorials are commoditized — there are 500 "how to build RAG" posts. What's scarce is honest accounts of what went wrong and what you learned. Failure posts work because they signal credibility (you've actually done the thing), create emotional resonance (everyone has failed), and provide non-obvious insight (the lesson is always more interesting than the success). The key is specificity: not "I failed at AI" but "I spent 40 hours on a fine-tuning approach that a 5-line prompt engineering change outperformed. Here's why."

2. Building in public

What to share vs what to protect

Share your process, your learnings, your mistakes, and your progress metrics. Protect client names, proprietary data, internal company strategy, and anything under NDA. The line is clearer than people think: "I built a RAG system that improved retrieval precision by 30%" is fine. "I built a RAG system for [Company X]'s internal compliance documents" is not. When in doubt, anonymize the context and keep the insight. The learning is what people follow you for, not the client name.

Week 1 progress posts outperform launch posts

Most people wait until a project is "done" to share it. By then, the audience has no context, no emotional investment, and no reason to care. Instead, share early: "Day 1: Starting a new project to build X. Here's my plan and what I'm nervous about." Week-1 posts outperform launch posts because people feel like they're part of the journey. They root for you. They share your updates. They remember you when you ship — because they watched you build it.

How to document a project without revealing proprietary work

Use the "pattern extraction" approach: instead of describing the specific project, describe the pattern you learned. "When building document processing pipelines, I've found that chunking strategy matters more than embedding model choice" teaches the same lesson without revealing anything proprietary. If you can describe the technical insight without mentioning the company, the data, or the business context, you're safe. Most insights are portable — the specifics aren't.

Turning build-in-public posts into inbound leads

Every build-in-public post should end with an implicit or explicit "and I do this for a living" signal. Not a hard sell — a positioning statement. "If you're working on a similar problem, happy to share what I've learned — DM me." or "I'm taking on 2 more consulting clients this quarter for exactly this type of work." The inbound comes from the combination of demonstrated expertise (the post) and availability (the CTA). One without the other doesn't convert.

3. Technical writing strategy

One deep post per month beats four shallow ones

For SEO and long-term credibility, depth wins. A 2,000-word post that thoroughly covers "How I reduced LLM hallucination rate by 60% using structured output validation" will rank on Google, get bookmarked, and get referenced in newsletters for months. Four 500-word posts covering surface-level topics will disappear in a week. Invest your technical writing time in fewer, deeper pieces. One definitive post per month builds a body of work that compounds over years.

The title formula that ranks

The highest-performing technical post title format is: "How I [specific action] by [specific percentage] using [specific tool]." Examples: "How I cut RAG latency by 70% using hybrid search with Pinecone" or "How I reduced LLM costs by 85% by switching from GPT-4 to fine-tuned Mistral." This format works because it promises a specific, achievable outcome, names the tools (which people search for), and implies a real-world implementation, not a theoretical tutorial. Write the title before the post — if the title isn't compelling, the post won't be either.

Where to publish: Medium vs Substack vs LinkedIn vs personal site

Medium: best for SEO reach if you get into a large publication (Towards Data Science, Better Programming). Downside: you don't own the audience. Substack: best for building a subscriber list with built-in discovery features. Good if you plan to write regularly. LinkedIn articles: lowest friction, good reach within your network, but poor SEO and not searchable outside LinkedIn. Personal site/blog: you own everything, best for long-term SEO, but zero built-in distribution. Best strategy: publish on your personal site for SEO ownership, cross-post to Medium or Substack for distribution, and share a summary on LinkedIn for immediate reach.

How to repurpose one technical post into 5 LinkedIn posts

Every deep technical post contains at least 5 standalone insights. Take each major section or finding and turn it into a standalone LinkedIn post with its own hook. Post 1: the main finding. Post 2: the biggest mistake you made along the way. Post 3: the counterintuitive insight. Post 4: the tool comparison. Post 5: the "if I had to do it again" reflection. Spread them over 2 weeks. Each one links back to the full post. This multiplies the return on every hour you invest in deep writing by 5x.

4. Newsletter launch

Beehiiv vs Substack — which to choose and why

Beehiiv is better if you want growth features: referral programs, recommendation networks, A/B testing subject lines, advanced analytics, and custom domains. It's built for newsletter operators who want to grow aggressively. Substack is better if you want built-in discovery: the Substack network drives subscribers to you through recommendations, and readers can find you through the app. Choose Beehiiv if you're a marketer at heart. Choose Substack if you want to focus on writing and let the platform handle distribution. For most people in AI, Substack is the faster start.

The issue format that retains subscribers

The highest-retention newsletter format for AI professionals is: 1 original insight (your unique take on a trend, tool, or technique — this is why people stay), 1 tool or resource recommendation (something practical they can use this week), and 1 industry signal (a funding round, product launch, or hiring trend that tells them where the market is going). Keep total length under 800 words. Readers unsubscribe from newsletters that take more than 5 minutes to read. Respect their time and they'll stay for years.

First 100 subscribers: DM your existing network

Don't post publicly about your newsletter until you have your first 50–100 subscribers. Why? Early issues need engagement (opens, replies, clicks) to build your sender reputation and avoid spam filters. DM 20 people per day for 5 days with a personal message: "Hey [name], I'm starting a weekly newsletter about [topic]. I think you'd find it useful based on your work in [their area]. Would you be interested?" Personal asks convert at 40–60%. Public posts convert at 1–2%. Get the foundation right first, then scale with content marketing.

How to monetize a small list before 1,000 subscribers

You don't need 10,000 subscribers to generate revenue. With 300–500 engaged subscribers in AI, you can: offer a paid tier with exclusive deep-dives ($10/month × 50 subscribers = $500/month), sell a digital product like a template pack or course ($49 × 20 buyers = $980), or use the newsletter as a lead magnet for consulting (one client from a newsletter is worth 1,000 free subscribers). The key is engagement rate, not list size. A 500-person list with 60% open rate is more valuable than a 10,000-person list with 15% open rate.

5. Speaking & conference strategy

Start with local meetups and virtual panels

One good talk generates 10x more inbound than 50 LinkedIn posts. But you don't start at KubeCon — you start at your local Python meetup, a virtual AI panel, or a company lunch-and-learn. These low-stakes environments let you practice delivery, test which stories resonate, and build a recording you can include in future CFP applications. Most people skip this step and wonder why their conference submissions get rejected. Organizers want speakers with a track record, even if it's three meetup talks.

CFP writing guide: submit case studies, not tutorials

Conference review committees see hundreds of "Introduction to RAG" submissions. What they accept is "How we reduced hallucination by 60% in our production RAG system — lessons from 6 months of iteration." The difference: case studies promise real-world insight, tutorials promise information anyone can Google. Your CFP should answer: what did you do, what happened (including what went wrong), and what should the audience do differently as a result? Include specific numbers, specific tools, and a specific outcome. Vague proposals get vague rejections.

Turn one talk into a 6-month content series

A single 30-minute conference talk contains enough material for: 1 blog post (the full written version), 4–6 LinkedIn posts (one per key insight), 1 newsletter deep-dive, 1 Twitter/X thread, and 1 YouTube video (the recording). Spread these across 6 months, referencing the original talk each time. This turns a one-time event into a long-tail content engine. Record every talk — even meetup talks — because the recording is the asset that keeps generating value long after the audience has left the room.

The follow-up system that converts conversations into opportunities

After every talk, you'll have 5–15 people approach you with questions or compliments. Most speakers say "thanks" and move on. Instead: ask for their LinkedIn, connect that night with a personalized note referencing the conversation, and follow up 3 days later with a relevant resource. This simple 3-step system (connect → personalize → follow up) converts 20–30% of post-talk conversations into ongoing relationships. Over a year of speaking, that's 50–100 warm connections in your target market — each one a potential client, collaborator, or referral source.