HACKATHON

PROBLEM STATEMENTS

AI Agents That Learn Using Hindsight

In this hackathon, you will build AI-powered applications using Hindsight, a memory system
developed by Vectorize that allows AI agents to remember, recall, and improve over time.

Instead of building AI that forgets conversations, your project should demonstrate persistent
memory and learning from past interactions.
Hackathon Problem StatementsAI Agents That Learn Using Hindsight

Why This Hackathon Matters for Your Career

The AI agent space is exploding, and companies are hiring people who can build, not just talk
about it. This hackathon is your chance to walk away with a working project that demonstrates
you can ship an AI agent that solves a real problem.
Here is what is at stake for you personally:

Build Your Portfolio, Not Just a Weekend Toy

The best hackathon projects don't just win prizes. They become the thing you link on your
LinkedIn profile, your GitHub pinned repo, and the story you tell in interviews for the next two
years. Pick a problem that a real business would actually pay to solve, and build toward that.

Learn by Doing What the Industry Needs Right Now

Stateless chatbots are yesterday's news. The frontier is agents with memory: agents that learn
from past interactions, adapt to user behavior, and get better over time. Hindsight gives you a
real memory layer to build on. That is a skill set that is immediately valuable.

Ship Something You Would Actually Use

The projects that stand out (to judges, to recruiters, to future collaborators) are the ones where
the builder clearly cared about the problem. Don't chase what sounds impressive. Build
something that solves a pain point you have actually experienced.

Required Technology
All teams must build their projects using Hindsight.

Official Links
• Hindsight Documentation: [Link]
• GitHub: [Link]
• Hindsight Cloud: [Link]

Resources
You can use Hindsight Cloud to quickly get a Hindsight instance, or the open source version.
Use promo code MEMHACK6 to get $50 in free credits on Hindsight Cloud. You add this
AFTER you register in the billing section.
Join the Hindsight Community Slack to ask any questions about how to use Hindsight.

LLM Access

Page 2
Hackathon Problem StatementsAI Agents That Learn Using Hindsight

You can use any LLM you want, but Groq is super fast and has a generous free tier:
[Link]
Recommended models are openai/gpt-oss-120b and qwen/qwen3-32b, but make sure to
have your agent ready to handle function calling errors.

Coding Agents
You can use any coding agent you want (or code by hand), but here are some free options:
• [Link]
• [Link]
• [Link] (can use with Ollama for local models or with most LLM providers)

OpenClaw Integration
If you want to build an agent using OpenClaw, you can use the official Hindsight plugin here:
[Link]

Page 3
Hackathon Problem StatementsAI Agents That Learn Using Hindsight

How to Pick a Winning Project

The difference between a forgettable hackathon demo and one that makes judges (and
recruiters) pay attention comes down to three things:

1. Solve a Real Business Problem

The strongest projects target a workflow that actual companies struggle with. Ask yourself:
"Would someone pay $50/month for this?" If the answer is yes, you are on the right track.
Bad: "An AI that remembers your favorite color."
Good: "An AI sales assistant that remembers every objection a prospect raised across calls and
drafts personalized follow-ups."
Think about the repetitive, knowledge-heavy tasks in real jobs. The ones where people say "I
wish someone would just remember all of this for me." That is your opportunity.

2. Make Memory the Star, Not a Feature

Since Hindsight memory accounts for 25% of the judging criteria, your project should make the
memory layer central and visible. The best projects will show a clear before/after: without
memory the agent is generic; with memory it becomes dramatically better.
Some ways to make memory shine:
• Show the agent improving its responses over multiple interactions
• Have it recall context from days or weeks ago
• Let it learn user preferences and adapt its behavior
• Have it build up domain expertise from past conversations

3. Think Like a Demo

You have limited time to impress judges. Pick a use case where the value is obvious within 60
seconds of watching. The best hackathon projects tell a story: "Here is the problem. Here is the
agent solving it. Watch how it gets smarter."

Page 4
Hackathon Problem StatementsAI Agents That Learn Using Hindsight

Project Ideas That Look Real

Below are categories of AI agents drawn from real business functions. Each one represents a
genuine workflow where persistent memory transforms the agent from a novelty into something
valuable.
Avoid student-centric projects like AI tutors, AI group project managers, AI quiz generators,
etc. Think about the professional world instead.
Browse this repository for more inspiration: [Link]

Sales & Revenue

Agent Idea Why Memory Matters Business Case
Deal Intelligence Remembers every interaction Sales reps waste hours re-reading
Agent across a deal cycle: objections CRM notes before calls. An agent
raised, competitors mentioned, with deal memory can brief a rep
stakeholder concerns, pricing in seconds and suggest winning
discussions. Over time, learns tactics based on past deals.
which objection-handling
approaches work best.
Outbound Learns which messaging angles Outbound is a numbers game, but
Prospecting Agent get responses from different personalization wins. An agent
persona types. Remembers what that remembers and adapts its
was sent to whom and what approach across hundreds of
worked. prospects is a force multiplier.
Proposal & RFP Builds a memory of past RFP responses are time sinks. An
Agent proposals, win/loss patterns, and agent that remembers your best
client-specific preferences. past answers and tailors them to
Recalls boilerplate that worked each new opportunity saves days
and language that didn't. per proposal.

Marketing & Content

Agent Idea Why Memory Matters Business Case
Content Strategy Tracks what content has been Marketing teams constantly
Agent published, what performed well, reinvent the wheel. An agent with
what topics have been covered, content memory can identify what
and what gaps exist. Learns your worked, what didn't, and where
brand voice over time. the gaps are.
SEO & Citation Remembers your site's ranking SEO is a long game. An agent
Agent history, past optimizations, that remembers the full history of
competitor moves, and what your optimization efforts and their

Page 5
Hackathon Problem StatementsAI Agents That Learn Using Hindsight

changes led to ranking outcomes can make smarter

improvements. recommendations than one-shot
analysis.
Social Media Learns which post styles, topics, Every brand's audience is
Engagement and timing drive engagement for different. An agent that learns
Agent your specific audience. YOUR audience's preferences
Remembers past conversations over weeks of interaction is far
and community sentiment. more valuable than generic
advice.

Engineering & DevOps

Agent Idea Why Memory Matters Business Case
Incident Response Remembers past incidents, their When production is down, every
Agent root causes, resolution steps, and minute counts. An agent that
which runbooks worked. Learns recalls exactly how similar
from post-mortems to suggest incidents were resolved before is
faster fixes for similar issues. invaluable.
Code Review Learns your team's coding Code review is slow and
Agent standards, common mistakes, and inconsistent. An agent that knows
architectural preferences over your codebase's patterns and your
time. Remembers past review team's conventions catches issues
feedback to avoid repeating the a generic linter never would.
same suggestions.
DevOps Pipeline Tracks deployment history, build An agent that remembers every
Agent failures, infrastructure changes, deployment failure and what
and their downstream effects. caused it can predict risks before
Learns which changes tend to they hit production.
cause problems.

Page 6
Hackathon Problem StatementsAI Agents That Learn Using Hindsight

Operations & Support

Agent Idea Why Memory Matters Business Case
Customer Support Remembers a customer's full Nothing angers a customer more
Agent history: past tickets, known issues, than repeating their story. An
their environment, their frustration agent with full customer memory
level, what solutions worked transforms the entire support
before. experience.
Accounts Payable Learns vendor patterns, payment AP teams process thousands of
Agent terms, common discrepancies, invoices and the same edge cases
and approval workflows. keep recurring. An agent with
Remembers past exceptions and memory handles exceptions that
resolutions. would normally require senior
staff.
Compliance & Tracks regulatory requirements, An agent that remembers your full
Audit Agent past audit findings, remediation audit history and can flag gaps
status, and policy changes. before auditors find them is
Remembers which controls have immediately valuable.
been tested and when.

Product & Strategy

Agent Idea Why Memory Matters Business Case
User Feedback Aggregates and remembers Product teams drown in feedback.
Synthesizer feedback from multiple channels An agent that connects dots
over time. Identifies emerging across channels and time spots
themes, tracks sentiment shifts, patterns humans miss.
and links feedback to specific
product changes.
Competitive Monitors and remembers Competitive intelligence is only
Intelligence Agent competitor moves: pricing useful if it is cumulative. An agent
changes, feature launches, that remembers six months of
messaging shifts, hiring patterns. competitor activity spots patterns
Builds an evolving picture of the that a weekly check never would.
competitive landscape.
Meeting Prep Remembers past meetings with An agent that briefs you with full
Agent each contact: what was context from every past interaction
discussed, what was promised, with a contact is a professional
what follow-ups were missed. superpower.
Learns your meeting style and
preparation preferences.

Page 7
Hackathon Problem StatementsAI Agents That Learn Using Hindsight

How to Make Your Project Stand Out

Tell a Story with Your Demo
Don't just show features. Walk through a realistic scenario: "Imagine you are a sales rep on your
fifth call with a prospect..." Make the judges feel the problem before you show the solution.

Show the Learning Curve

The most compelling Hindsight demos show the agent getting noticeably better. Interaction 1:
generic. Interaction 5: personalized. Interaction 20: feels like it knows you. That progression is
what makes memory-powered agents exciting.

Build Something You Would Put on LinkedIn Tomorrow

If you wouldn't proudly post about it, rethink your approach. The best hackathon projects get
written up as case studies, attract followers, and start conversations. Aim for that.

Keep the Scope Tight

You have limited time. A polished agent that does one thing brilliantly beats a sprawling
prototype that does five things poorly. Pick one workflow, one persona, one clear value
proposition and nail it.

Use Realistic Data

If you can find real data on sites like Kaggle or HuggingFace that’s great. Even if you can’t, use
your LLM to generate synthetic data that looks real. Give your customers real-sounding names,
your deals real-sounding numbers, your incidents real-sounding error logs.
The #1 thing that will make your project look real is the data, so don’t make sure you give this
the attention it deserves.

Important Rules
All teams must ALSO share their project based on challenges from the official content guide.
Content Guide: Hackathon Content Guide (see official link)
Your project must clearly demonstrate how Hindsight memory is used in your solution.

Submission Requirements
Each team must submit:
• GitHub Repository with clean, documented code
• Demo Video showing the agent in action
• Live Project Demo to judges

Page 8
Hackathon Problem StatementsAI Agents That Learn Using Hindsight

• Content deliverables: All team members must complete an Article, Social Media post,
and Video as described in the content guide
• Explanation of how Hindsight memory is used in your solution

Judging Criteria
Criteria Weight What Judges Are Looking For
Innovation 30% Is this a fresh take on a real problem? Does it go beyond obvious
chatbot territory?
Use of 25% Is memory central to the value proposition? Does the agent
Hindsight clearly improve over time?
Memory
Technical 20% Is the code clean, well-architected, and functional? Does it
Implementatio handle edge cases?
n
User 15% Is the agent intuitive to interact with? Does the demo flow tell a
Experience compelling story?
Real-world 10% Could this solve a genuine problem for real users? Is there a
Impact path to actual adoption?

Now stop reading and start building.

The best way to learn AI agents is to ship one.

Page 9