Forge AI Hackathon · April 17–18, 2026
AI for
First-Timers.
Never used AI tools before? You're in the right place. This guide will get you from zero to building in under 10 minutes. No coding experience required.
Start at claude.ai
Use with Lovable to build apps
Credits pre-loaded for all teams
Section 01
What even is AI?
Think of an AI assistant as a very capable colleague who never gets tired, never judges you, and knows a lot about a lot of things. You talk to it in plain English - it reads, writes, analyses, codes, and thinks alongside you.
✍️
It writes
Emails, proposals, cold outreach, policy docs, summaries, reports - anything that needs words. Give it context and it drafts for you.
→ Try it first
🧠
It thinks
Stuck on a problem? Paste it in. It will break it down, poke holes in your plan, suggest angles you haven't thought of, and reason through tricky decisions.
→ Great for ideation
💻
It codes
Even if you've never written code, it can build working web apps, scripts, and tools when you describe what you want. This is how you'll use it with Lovable.
→ Power move for builders
The one thing to remember
The AI is only as good as what you tell it. A vague question gets a vague answer. A specific, well-structured question gets something genuinely useful. The rest of this guide is about how to ask well.
Section 03
Rules of the game
Six things that will make you immediately better than 80% of AI users - and six traps beginners fall into.
✓ Do these
01
Paste real data. Drop in a real email, a CSV row, a policy paragraph. It responds to your actual content, not a hypothetical. Specificity is everything.
02
Iterate in the same chat. Never start over. Just say "make it shorter", "add urgency", "change the tone to formal". It remembers the full conversation thread.
03
Give it a persona. "You are a senior CS manager at a health insurance company…" - this single line massively improves relevance and quality.
04
Ask it to think step-by-step. For complex tasks, add "think through this step by step before giving me the answer." Forces deeper reasoning.
05
Use it as a rubber duck. Stuck? Paste your half-baked idea and ask "what am I missing?" or "poke holes in this plan."
06
Ask for options. "Give me 3 different versions - one casual, one formal, one urgent." Then pick what works and ask it to refine it.
✕ Avoid these
01
Don't be vague. "Write me something about insurance" → generic output. The more context you give, the better the result. Always.
02
Don't accept the first output. The first response is always a draft. Push back: "That's too corporate - make it warmer." "Too long - cut by 40%."
03
Don't forget the audience. It doesn't know if you're writing for a CFO or a fresh grad unless you say so. Always specify who will read this.
04
Don't treat it like Google. AI isn't a search engine. It reasons and creates - use it for thinking, drafting, structuring, building. Not for "what's the capital of France".
05
Don't start a new chat for every change. You lose all context. One long conversation with iterations beats five fresh chats every time.
06
Don't skip the "why". Tell it why you need something. "Because we want to reduce churn" gives it direction to make smarter choices autonomously.
Section 04
When it goes wrong
It won't always nail it on the first try. That's normal. Here's what to say to steer it back - no need to start over.
Say these exact phrases
→
Too generic? "This could be about any company. Rewrite it specifically for Nova Benefits, whose audience is HR managers at mid-size Indian tech firms."
→
Too long? "Cut this by half. Keep only the most important parts. The final version should be under 100 words."
→
Wrong tone? "This sounds too formal. Rewrite it like a peer talking to a peer - warm, direct, human."
→
Off-track? "You misunderstood - I meant X, not Y. Please try again with that correction."
→
Making things up? "Only use the information I've given you. Do not invent facts, names, or numbers."
→
Too safe / hedging? "Stop adding disclaimers. Give me a direct, confident answer."
Phrases you can add to any prompt
→
"Think step by step before answering."
→
"Give me 3 versions: casual, formal, urgent."
→
"Point out any assumptions you're making."
→
"If you're unsure about something, say so - don't guess."
→
"Summarise this into 5 bullet points."
→
"Respond as if writing for a non-technical audience."
→
"What questions would you ask me to do this better?"
Section 05
Real prompts, mapped to your problems.
Example prompts for the kinds of problems you'll encounter at the hackathon. Use these as a reference - notice how each one applies the Role + Task + Format formula.
Cold Email Agent · Problems #1 & #3
✓ Good prompt
You are a BDR at Nova Benefits, a corporate health insurance platform in India. Your ICP is HR Heads at tech companies with 100–500 employees.
Write a cold email to the Head of People at a Series B SaaS startup. The company recently expanded from 80 to 200 employees. Nova's key edge: claims processed in 48 hours vs. industry average of 2 weeks.
Goal: book a 20-min intro call.
Tone: warm, peer-to-peer, not salesy.
Length: under 120 words. End with a low-friction CTA.
HR Policy Chatbot · Problem #6
✓ Good prompt
Here is Nova Benefits' leave policy:
[paste your actual policy text here]
An employee just asked: "Can I take 2 days of sick leave without a doctor's certificate?"
Answer the question based strictly on the policy above. Cite the exact section. If the policy doesn't cover this, say so clearly and suggest they contact HR.
Competitor Research Brief · Problem #7
✓ Good prompt
I'm a BDM at Nova Benefits preparing for a sales call where the prospect currently uses Plum Health.
Create a structured competitive brief on Plum with:
1. Product overview
2. Known pricing signals
3. Common objections Nova hears about Plum
4. 3 talk-track responses Nova can use
Format as a quick-reference table + 3 bullet responses I can memorise.
Accounts Receivable Follow-Ups · Problem #5
✓ Good prompt
You manage AR at Nova Benefits. Write 3 escalating follow-up emails for an overdue invoice:
Email 1 (Day 3): Gentle reminder, assume it slipped through.
Email 2 (Day 10): Firmer, mention impact, request confirmation date.
Email 3 (Day 21): Formal notice, CC the AM, give 5-day hard deadline.
Tone: professional throughout, never rude.
Each email under 100 words.
✦ Example prompt to generate a Lovable app spec
✦ Example
I'm building [describe your problem statement from the hackathon brief].
Help me write a product spec for Lovable (a no-code app builder) that includes:
1. What the app does in one sentence
2. The main user and their goal
3. Key screens / views needed
4. What inputs the user provides
5. What Claude (AI) should do in the background
6. What the output looks like
Keep it concise - Lovable works best with clear, direct instructions.
Section 06 · For App Builders
System Prompts
A system prompt is the hidden instruction you give the AI before any user talks to it. When you build an app in Lovable, this is how you define the AI's personality, scope, and rules - so every user gets a consistent, on-brand experience. Think of it as the briefing document the AI reads before your app goes live.
System Prompt - HR Policy Chatbot
You are Nova Benefits' internal HR assistant.
RULES:
- Answer ONLY using the policy documents provided below.
- Always cite the specific section you're drawing from.
- If you're less than 80% confident, say "I'm not certain - please confirm with HR at hr@nova.com."
- Never make up policies. Never guess.
- Keep answers under 150 words unless asked for more.
- If the question is about payroll, redirect to finance@nova.com.
TONE: Friendly, clear, helpful. Not robotic.
[POLICY DOCUMENTS WILL BE INSERTED HERE]
System Prompt - Competitor Intel Agent
You are a competitive intelligence assistant for Nova Benefits' sales team.
When given a competitor name, produce a structured brief with:
1. Company overview (2–3 sentences)
2. Core product features
3. Pricing signals (public info only)
4. Weaknesses / gaps vs Nova
5. 3 talk-track bullets for Nova BDMs
ALWAYS end with: "Last updated: [today's date]. Verify recent news before a live call."
TONE: Sharp, direct, sales-ready. No fluff.
FORMAT: Use headers. Full brief under 400 words.
Section 07
Connecting AI to real data
The AI on its own is powerful. With access to your actual files, emails, and databases is a different thing entirely. Here's how to wire it up - from no-code to full API.
Connectors claude.ai
Pre-built integrations inside claude.ai. Connect Google Drive, Gmail, Slack, Jira in one click. It can then read your actual files and emails - no code needed.
Where: claude.ai → Settings → Connectors
Use for: "Summarise the Q3 report in my Drive" or "Draft a reply to this Gmail thread"
MCP Servers advanced
Model Context Protocol - lets the AI talk to any tool or database in real-time. Used in custom apps built with the API. Think of it as the AI's plugin system for developers.
Use when: building a custom app that needs the AI to read live data from your CRM, invoicing tool, or internal database.
Anthropic API Lovable
The API lets your Lovable app call the AI programmatically - so your app can generate emails, analyse data, or answer questions automatically, without a human typing prompts.
Lovable handles this for you. Just describe what you want the AI to do inside your app and Lovable will wire it up.
How to use Connectors in claude.ai (no code needed)
03
Connect Drive / Gmail / Slack
05
Ask about your real data
✓ With Google Drive connected
"Read Nova_Policy_Benchmarks.xlsx in my Drive and tell me which parameters are RED for a 300-person IT company."
✓ With Gmail connected
"Find all emails from clients with the word 'invoice' in the last 30 days. Summarise which ones haven't received a reply."
Calling the AI from your Lovable app - the key snippet
Lovable generates this for you, but understanding the shape helps you brief it better. Three things matter: your system prompt (rules), the user message (input), and the model.
// Inside your Lovable app - Claude API call
const response = await fetch("https://api.anthropic.com/v1/messages", {
method: "POST",
headers: { "Content-Type": "application/json" },
body: JSON.stringify({
model: "claude-sonnet-4-20250514", // always use this model
max_tokens: 1000,
system: "You are Nova's AR assistant. Only write follow-up emails...",
messages: [{ role: "user", content: userInput }]
})
});
const data = await response.json();
const claudeReply = data.content[0].text; // Claude's response
✦ Building a system? Ask the AI to return JSON
When the AI is a step inside a larger app or workflow - not just answering a human - prose output is hard to use. Instead, tell it to return structured JSON. Your app can then read the fields directly without any guesswork or text parsing.
✕ Prose output (hard to use in code)
Analyse this invoice email and tell me the client name, amount, and whether it's overdue.
Result: "The client is Acme Corp, the amount due is ₹45,000, and yes it appears to be overdue by 12 days." - you'd need to parse this string to use it.
✓ JSON output (plug straight into your app)
Analyse this invoice email. Return ONLY a JSON object with these fields: client_name, amount_inr, due_date, is_overdue (boolean), days_overdue (number or null). No explanation, no prose - just the JSON.
Result: {"client_name":"Acme","amount_inr":45000,"is_overdue":true,"days_overdue":12} - ready to use.
Section 08
AI → Lovable workflow
The most powerful pattern in this hackathon: use AI to think and spec, then use Lovable to build. Don't open Lovable without a clear spec first.
✦ Pro move - write your requirements with AI before you build anything
Before telling any AI agent (or Lovable) to start building, take 10 minutes to make your requirements crystal clear. Vague input → broken output. The AI is excellent at turning a rough idea into a tight requirements doc - and the better your spec, the less time you waste fixing things that shouldn't have been built that way in the first place.
✦ Requirements prompt
I want to build [describe your idea in 2–3 sentences].
Help me write a clear requirements document before I start building. Include:
1. Problem statement - who has this pain, and what exactly is it?
2. Core user flow - what does the user do, step by step?
3. Must-have features for a working MVP (be ruthless - only what's essential)
4. Out of scope - what are we explicitly NOT building?
5. Success criteria - how do we know it's working?
Be specific. Flag any assumptions you're making.
04
Paste spec into Lovable
Step by step
01
Pick a problem statement from the hackathon brief. Read it carefully. Write down in one sentence: what is the pain and who feels it?
02
Open claude.ai and paste the spec generator prompt (Section 05 above). Fill in your problem. It will return a clear, structured product brief.
03
Review and refine the spec. Ask: "What's the simplest version of this I can build in 24 hours?" Push it to be concrete and achievable.
04
Open lovable.dev and paste your spec. Lovable will scaffold the UI and wire up the AI calls automatically. You're building, not coding.
05
Stuck in Lovable? Come back to the AI tool. Paste the broken code or the error message and ask: "What's wrong and how do I fix it?"
AI + Lovable tips
✓ Stuck in Lovable?
"Lovable generated this component but the button doesn't trigger the email send. Here's the code: [paste]. What's wrong and how do I fix it?"
✓ Need better UX?
"I'm building an invoice tracker dashboard. What are the 5 most important things a finance manager needs to see at a glance? Give me a layout suggestion."
✓ Need a demo script?
"My app does X for Y audience. Write me a 5-minute demo script for a panel of judges who care about business impact."