Pitch + capability map

This is the verbal OS for the whole Principal search path — war room through peer seat. Same brand every time. Later chapters add war stories. They do not get a new identity.

Brand: I design and ship governed agent platforms — orchestration · governance · access-aware RAG · eval gates. Not chat demos.

Honesty: Lucid outcomes are employer production. Public spine (AegisAI, VAP, Enterprise RAG, Content Factory, golden-eval) is inspectable proof of the same principles. I never claim those repos ran inside Lucid.

Why I keep one pitch

Recruiters hear keywords. Hiring managers hear ownership. Staff+ / Principal panels hear judgment under irreversible risk. One pitch has to survive all three without becoming a catalog tour.

Five beats (memorize the beats, not every comma)

If they interrupt, I finish the current beat, then ask their problem.

What I actually say (panel / HM open)

I design and ship governed agent platforms — not chat demos, and not agent builders without a control plane.

Four decisions have to stay explicit. Orchestration decides what agents do. A governance plane decides what they’re allowed to execute. Access-aware RAG decides what they’re allowed to know. Eval gates decide whether we ship. Miss any of those and you don’t have a platform — you have a demo that can write to production.

At Lucid I took enterprise AI from 0→1 on a real cost problem: 20,000+ suppliers stuck on TrueCommerce EDI. We replaced the vendor with three LangGraph orchestrators — router to 34 specialists, shared graph state plus RAG over vector stores, Kafka as the event backbone to the microservices. Onboarding, releases, shipments, plus a supplier portal. That’s $10M licensing eliminated and $7M annual net after we run it ourselves. Kafka/MSK ownership was also a $1M+ infra save. Payments unlocked EU and Gulf rails. Production inside a governance envelope, not a lab toy.

Publicly I keep a spine a panel can actually click. The two I’d go deep on are governance — gateway, HITL, audit — and access-aware RAG, authZ before ranking. Same principles as the Lucid work. Not a claim those binaries ran inside Lucid.

If it’s useful, a short technical review — or we pick one of those two and go deep against the problem you’re hiring for.

90-second cut (cold)

I ship governed agent platforms, not chat demos. Orchestration decides what agents do; a control plane decides what they’re allowed; access-aware RAG decides what they can know; evals decide whether we ship. At Lucid I replaced a $10M TrueCommerce EDI lock-in with LangGraph routers, shared RAG over vector stores, and Kafka to the microservices — 20k suppliers, $7M net. Public proof you can poke: governance and access-aware RAG. Same principles, not a claim those repos ran inside Lucid. Happy to walk a 15-minute review or go deep on whichever matches your problem.

45-second HM screen

I ship governed agent platforms, not chat demos — control plane, access-aware RAG, evals that can fail CI. At Lucid I took enterprise AI 0→1: $10M TrueCommerce licensing off the table, $7M net. Nineteen years on platforms. If useful: venkat-ai.com/hire. I’m looking for Sr Staff / Principal / Founding where agents and retrieval are the actual product.

30-second recruiter

I ship governed agent platforms, not chat demos. Lucid 0→1 supply-chain AI — $10M licensing eliminated, $7M net. Open to Sr Staff / Principal / Founding. Proof: venkat-ai.com/hire.

15-second DM

Building governed agent platforms, not chat demos. Lucid 0→1 supply-chain AI — $10M licensing gone, $7M net. Technical review: https://venkat-ai.com/technical-review

LinkedIn headline

Principal AI Architect · Agent Governance · Multi-Agent Platforms · RAG & Evals · Lucid · Open to Sr Staff AI Platform / Founding

Outreach blurb

I’m exploring Senior Staff / Principal AI platform and Founding roles focused on governed agent systems. At Lucid I replaced TrueCommerce EDI with an in-house multi-agent platform — $10M licensing eliminated, $7M annual net — plus Kafka/MSK and multi-rail payments. I also ship a public reference stack: gateway, orchestration, access-aware RAG, eval CI. Happy to share a 15-minute technical review: https://venkat-ai.com/technical-review

Delivery

Slight pause after “not chat demos.” Slow on the four planes. Numbers once. Don’t mash $10M licensing + $7M net + $1M Kafka into one vanity total. Don’t re-explain unless they ask.

If they interrupt: finish the sentence, then:

Happy to go deeper — what’s the problem this role is solving in the next 6–12 months?

I do not restart from beat 1.

“So you’re a LangChain guy?”

Frameworks are implementation. The architecture is control planes: what agents may do, what they may know, and what can fail a build.

“Is this production at Lucid or open source?”

Lucid is employer production with outcomes I can defend. The public spine is the inspectable reference for the same principles. Useful for panels. Not a claim those exact services ran inside Lucid.

“We just need a ChatGPT wrapper / LangChain engineer.”

I can ship a thin wedge. What I won’t do is put an irreversible tool behind a prompt with no control plane. If refunds, CRM writes, or publish are in scope, we talk gateway and HITL on day one — otherwise we’re demoing, not deploying.

“ChatGPT already does this.”

ChatGPT doesn’t know your tenant boundary, and it won’t refuse a CRM write. Side effects and authZ-before-rank are the product. Happy to walk a deny path instead of debating the chatbot.

“Are you still at Lucid?”

Yes. Nov 2023 through present. I’m looking, confidentially. I don’t broadcast that inside the company.

“Why this company?”

I did the homework on [team / problem — not the logo]. I want a seat where agents and retrieval are the product, with a control plane around tools that can do damage. If this loop is a ChatGPT wrapper with a Staff sticker, we’re misaligned.

“Walk me through a project.”
STAR 1 in 03 — wedge, then specialists under a control plane. I don’t start with headcount.

“What are you looking for in comp / start date?”

I’m scoring Bay Area Sr Staff AI total-comp, not a Senior band with an AI title. Exact numbers I’ll do with recruiting. Start date is a professional notice at Lucid — I won’t ghost a team to look hungry.

I never say: “Seventeen platforms.” “Synergies.” “Agentic transformation.” “Stepping stone.” Fake SLOs. Cursor ghost-authorship stories. I don’t name VAP, Content Factory, and golden-eval in the open unless they pull the thread. Two heroes. Rest on request.

Record once. If it sounds like a brochure after find-replace of my name, I add a deny path or a limitation.

Capability map — every JD → two or three rows

Pick two or three. Offer one deep dive.

FDE / Applied JD: same map. Deep dive is discovery → wedge → HITL, not a second pitch.

Call flow (same every time)

Pitch
  → Their problem (6–12 months)
  → Map → 2–3 rows
  → Proof (review + ONE deep dive)
  → Close (stage · timeline · 90-day success)

1 — Pitch

Full or 90s. Stop after the invite. Don’t ask “want me to walk all five demos?”

2 — Their problem (the question that actually matters)

What problem is this role solving in the next 6–12 months — for the team and for whoever owns the P&L?

Listen for: irreversible tools, multi-tenant data, RAG quality, cost, founding wedge, “ChatGPT wrapper” skepticism, Java platform politics.

If they dodge (“we’re transforming with AI”):

Happy to work with a fuzzy charter. What’s the riskiest action an agent could take in your world today — refund, publish, CRM write, deploy, or something else?

3 — Map out loud

That sounds like primarily [row A] and [row B] — with [row C] as the constraint. Here’s how I’d show proof without a catalog tour…

One sentence per row. Cap at three. If they named six themes, pick the two that touch irreversible risk or false confidence.

4 — One deep dive

Two ways to go deep: a short technical review of the live spine, or a whiteboard on governance, access-aware RAG, or multi-tenant isolation — whichever matches what you just described. I won’t spray five demos.

5 — Close

Where is this loop, what’s the decision date, and what does a good hire look like in the first 90 days? I want to respect your process — and I don’t want to guess.

Capture offline: stage · timeline · 90-day success · next-step owner · governance vs RAG.

Recruiter with no substance: still ask timeline and level band. Offer hire + technical-review links. Don’t dump STARs.

Light keyword echo only. Guardrails → governance. Multi-agent → orchestration. RAG → RAG. Eval → eval. 0→1 → founding. Java/Go → honesty + enterprise depth. Brand does not change.

How the same pitch ages across the arc

I rehearse this page every week. I do not fork an Anthropic pitch and a FAANG pitch.

Weekly rehearsal

1. Full pitch aloud once. Record every other week.
2. Call flow with a timer: they answer step 2 with a fake JD; I map two or three rows.
3. Interrupt drill: they cut me at beat 2 — I recover into their problem.
4. Practice Arena when useful.

Done when: 90s cold. Any JD → ≤3 rows in under a minute. One deep dive without listing the catalog. Close with stage / timeline / 90-day success every time.