Talks And Proposals

Talks & Proposals Skill

Notion Database

DB: CFP (Tech talks, Workshops and Papers) URL: https://www.notion.so/hj1393/4132a63bb32a4d93a724aeb027f92077 Collection ID: collection://767f10ea-c283-4294-8c86-0c618651ce5d

Schema:

• Proposal (title) — the talk/paper/workshop title
• Program (select) — the specific conference or program:

JOSS · Linux Foundation · Grant · Conference CFP · Other · ICML 2026 · MCP Conference · AI Engineer Fair 2026 · MLSys 2026 · MLOPS World 2026 · KubeCon2026 (add a new option when submitting to a conference not yet listed)

• Type (multi-select) — Talk · Workshop · Demo · Tutorial · Paper
• Status (status) — full pipeline:

To draft → Drafting → Pending submission → In review → Submitted → Accepted / Rejected → Delivered

• Deadline (date) — submission deadline
• Owner (person) — assign to Harshul
• Evidence (relation) — link to EB1A evidence entries once accepted

Always set at minimum: Proposal, Program, Type, Status, Deadline.

---

About Harshul

Senior ML Engineer at Audible (Amazon), ~9 years. Builds GenAI search, real-time ML feature stores (100K+ QPS), fraud detection. Creator of AgentShip. MS CS NYU.

His canonical talk reference: "Building a Real-Time ML Feature Store at 100K QPS" — use this as the quality bar for specificity, grounding, and what "production experience" sounds like in a proposal.

---

Task 1: Draft a CFP Submission

Produce a complete, paste-ready CFP. Structure:

For a Tech Talk (30–40 min)

TITLE
[Specific + outcome-forward. Max 10 words. The title should hint at the insight, not
just describe the topic.]

❌ "Introduction to ML Feature Stores"
✅ "Building a 100K QPS Feature Store: What Nobody Warns You About"

---
ABSTRACT (~200 words)

[Para 1 — The problem, made real. 2-3 sentences. Something a senior engineer
has actually felt.]

[Para 2 — What you'll show. Ground it: name the system, name the scale, name Audible
if relevant. "In this talk, I'll share how my team at Audible (Amazon) built X to
handle Y. I'll walk through [specific decision], [real tradeoff], [real lesson]."
No fluff — if a sentence doesn't add information, cut it.]

[Para 3 — Concrete takeaways. Not "you'll learn about X." Instead: "You'll leave
with [specific thing] you can apply to your own system on Monday."]

---
TALK OUTLINE

[00:00–05:00] Problem Space — [1-sentence description]
[05:00–15:00] Architecture Deep Dive — [description]
[15:00–23:00] War Stories / Incidents — [what went wrong + lesson]
[23:00–33:00] North Star / Vision — [where is this going]
[33:00–38:00] Key Takeaways — [3 concrete rules]
[+ 10 min Q&A]

---
TARGET AUDIENCE
• [Persona 1 — who they are and why this talk matters to them]
• [Persona 2]
• [Persona 3]

---
SPEAKER BIO
Harshul Jain is a Senior ML Engineer at Audible (Amazon), where he builds real-time
ML infrastructure including a feature store serving 100,000+ queries per second.
With 9 years of production engineering experience, he specializes in GenAI-powered
search, fraud detection, and ML systems at scale. He created AgentShip, an open-source
AI agent deployment framework, and mentors 300+ engineers on Codementor. He writes
The Engineer's Digest on Substack.
LinkedIn: [URL] | GitHub: [URL]

---
FORMAT: Talk (35 min + 10 min Q&A)
A/V: HDMI, terminal display, mic

For a Workshop (60 min)

Same header structure, but outline becomes:

[00:00–05:00] Hook — what participants will have built by end
[05:00–15:00] Concept I — live walkthrough
[15:00–25:00] Exercise 1 — guided, everyone finishes
[25:00–35:00] Concept II
[35:00–55:00] Exercise 2 + stretch challenge
[55:00–60:00] Wrap-up + 3 takeaways

Prerequisites: [list clearly]
Max participants: 25–30
Materials: GitHub repo with starter code, setup guide sent 48h before

For a Paper / JOSS Submission

TITLE

ABSTRACT (~250 words)
[Para 1 — Motivation: what problem does this solve, who has it]
[Para 2 — What the software/paper does: technical approach]
[Para 3 — Novelty vs. existing work: why this matters, what's different]
[Para 4 — Usage and impact: who's using it, what metrics show adoption]

STATEMENT OF NEED
[Why this tool/paper fills a gap in the field]

TARGET VENUE: [JOSS / ArXiv / conference workshop track]

---

Task 2: Tailor for a Specific Conference

When the same talk goes to multiple conferences, adjust:

| Conference type | Tone | Emphasize | |---|---|---| | Practitioner (QCon, SREcon, InfoQ) | Peer-to-peer, blunt | What went wrong, real numbers, architecture | | ML/AI (MLOps World, Applied ML) | Technical, metrics-forward | Scale, production vs. benchmark, operational lessons | | Broad tech (AWS re:Invent, KubeCon) | Slightly accessible | Business outcome first, then mechanism | | University / hackathon | Energetic, encouraging | Career advice + technical depth |

The abstract changes. The talk itself doesn't.

---

Task 3: Log to Notion

After drafting, always create the entry in Notion immediately. Use notion-create-pages:

Database: collection://767f10ea-c283-4294-8c86-0c618651ce5d

Properties to set:

Proposal:  [Talk/Paper/Workshop Title]
Program:   [Conference name from select list, or "Conference CFP" if new]
Type:      ["Talk"] or ["Workshop"] or ["Demo"] or ["Tutorial"] or ["Paper"]
Status:    "To draft"  (move to "Drafting" once you start writing)
Deadline:  [Submission deadline date]

Page body: Full abstract, talk outline, speaker bio, submission notes. Update Status as the proposal moves through the pipeline — this drives the Pipeline board view.

Status lifecycle: To draft → Drafting → Pending submission → In review → Submitted → Accepted or Rejected → Delivered

Once accepted, link the Evidence field to the corresponding EB1A evidence entry.

---

CFP Abstract Quality Check

Before submitting, verify:

• ✅ First sentence makes a non-obvious claim (not a description)
• ✅ At least one specific metric or system reference by paragraph 2
• ✅ Company/system named (Audible/Amazon when relevant — signals credibility)
• ✅ Closing sentence says exactly what attendees walk away with
• ❌ No "journey", "explore", "gain insights", "fascinating", or "overview of"
• ❌ No vague outcomes ("better understanding", "new perspective")

---

EB1A Note

Every accepted talk = Criterion 5 (Original Contributions) + Criterion 7 (Critical Role). After a talk is accepted, capture: conference name, attendance, acceptance rate, invitation email, slide deck, recording URL. Add these to the Notion page body.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: harshuljain13
• Source: harshuljain13/agent-skills
• License: MIT

Install and usage instructions live in the source repository linked above.