I have been doing open-source for 10 years now. From side projects to small and popular libraries and to hackathon projects.
That's why I when I saw the DEV's Hacktoberfest Weekend Challenge. I knew I had to jump on it and share some ideas with what you can build with Valyu as part of your toolkit for building an awesome project!
You have until October 5 to make it useful to somebody you actually know, with open-source AI at its core.
I'd start with a favor you already do for people. Perhaps you help a friend research companies before interviews, or you keep checking scholarship requirements for a relative. You already understand a bit of the work. Ask them to show you the tedious part, then build around that.
I've put together 20 ideas below, along with the open source apps I've built (or contributed to), code you can learn from, and advice for writing the submission. Pick the one that makes you think of a particular person or anyone that resonates well with you!
First, the challenge in plain English
The theme is Build for a Friend. Pick one real person, understand a problem they actually have, and ship a new project during the challenge window.
According to the challenge page and contest-specific rules:
| What matters | The detail |
|---|---|
| Build window | October 2, 2026 at 2:00 AM UTC to October 5, 2026 at 6:59 AM UTC |
| Deadline | October 5 at 6:59 AM UTC; October 4 at 11:59 PM PDT |
| Core requirement | Open-source AI must make the project work: an open-weight model, an open-source agent harness or framework, local inference, or a combination |
| Submission | A published DEV post using the official template, a deployed demo or video, a code link, and an explanation of why open innovation matters |
| Judging | Writing quality is weighted most heavily; the five published criteria are discussed below, and the challenge pages do not specify numeric weights |
| Cash prizes | $2,450 across 17 winners: one $250 overall prize, six $200 featured-category prizes, and ten $100 partner-category prizes |
| Other awards | The overall winner also receives DEV++ membership; winners receive an exclusive badge, and valid participants receive a completion badge |
| Teams | Up to four people; one submission per team, with teammates' DEV handles in the post |
You can enter one submission in every partner category it genuinely qualifies for, but it can win only once in this challenge. You do not need a partner integration to win the overall prize.
This edition asks for new projects, rather than a pull-request count. Existing open-source work can inform your build, but a renamed old app is not a new submission. We'll come back to reuse and attribution.
Where AI Search fits, and where the open part fits
Most interesting friend problems have two halves.
The first is finding out what the problem is. The second is making the information at your disposal usable for the person or a group of persons. Your friend needs a comparison sheet, an explanation in their language, an interview rehearsal, or a checklist they can follow on their phone.
That gives you a practical architecture:
Friend's question or task
|
v
Your app / open-source workflow
|
+--> Valyu Search + Contents: retrieve and read evidence
|
+--> Valyu DeepResearch: investigate a larger question
|
v
Open-weight model: explain, compare, personalise, or rehearse
|
v
A useful output with source links
Which Valyu capability should you use?
| Your app needs to… | Use… | A friend-sized example |
|---|---|---|
| Find current information | Search | Find scholarships matching a subject and country |
| Read specific pages | Contents | Extract eligibility and deadlines from official pages |
| Get a search-grounded answer | Answer | Produce a quick company briefing with citations |
| Investigate a multi-part question | DeepResearch | Compare three study destinations, costs, and application requirements |
For the clearest open-model-centered build, start with Search and Contents, then let your open model do the synthesis.
If you use Answer or DeepResearch, give the open component a substantial downstream job: an interactive tutor, a comparison workflow, a source-review agent, or a personalised action plan.
Valyu covers the web alongside academic, financial, patent, and life-sciences sources. Check docs.valyu.ai/guides/datasources before promising a dataset: specialised sources have plan requirements, and some families, including transportation, environmental feeds, and automotive records, are available through DeepResearch rather than raw Search.
20 projects you can build for someone you actually know
Each idea below depends on external information evidence. Valyu supplies that evidence; the open component makes it useful for your friend.
The names are suggestions. The people are illustrative. Replace the persona with someone you know, and keep the first version small enough to finish.
1. Deadline Buddy: a scholarship shortlist for your sister
Your sister spends hours reading scholarship pages, only to discover she is ineligible halfway through an application.
Build a shortlist for one subject, destination, and study level. Use Valyu Search to discover opportunities and Contents to extract requirements from official scholarship pages; Compare those requirements with her profile and explain each match or mismatch.
Ship five opportunities with deadlines, funding details, evidence links, and a "needs clarification" label. The demo moment: an attractive scholarship is rejected for a real eligibility reason, and she can click through to the exact requirement.
2. Interview Espresso: a rehearsal grounded in the actual company
Your friend has an interview on Monday and knows the job description better than the company.
Use Valyu Search and Contents to research the employer's products, recent announcements, and the role. Let an open model run a mock interview based on those findings and your friend's locally stored experience, then explain where an answer needs more detail.
Limit the build to one company and one role. Generate a one-page brief and five practice questions. Demonstrate a question tied to a recent announcement, with its source, rather than another generic "What is your biggest weakness?"
3. Paper Compass: a reading queue for your PhD friend
A friend is starting a research topic and cannot tell which papers deserve their weekend.
Give Valyu a narrow academic question and retrieve relevant papers. Use an open model to organise them by research question, method, findings, and limitations, with a clear distinction between abstracts and available full text.
Start with ten papers and an editable comparison table. Your demo should show one disagreement between studies and link to the passages behind it. A useful reading queue admits uncertainty; it does not invent a tidy consensus.
4. Trial Translator: a research brief for a caregiver
Someone you love is trying to understand what researchers are testing for a condition.
Use Valyu to retrieve clinical-trial records and related literature. An open model explains trial phase, recruitment status, locations, and published eligibility language in ordinary English, then drafts questions to take to a clinician.
Scope the first version to one condition and region. Show a source-linked shortlist with trial IDs and verification dates. A strong demo exposes an eligibility criterion the app cannot assess, rather than claiming a person qualifies for a trial.
5. Supplement Receipts: a Claim truth checker for your gym friend
Your friend has sent you another video insisting a supplement will fix everything.
Use Valyu to find human studies and reviews about one specific claim. Let an open model distinguish study populations, outcomes, limitations, and evidence gaps, then produce a claim-versus-evidence card.
Begin with one supplement and three claims. Show where a marketing sentence stretches beyond the study it cites. Keep the output focused on evidence and questions, rather than personalised dosing or treatment recommendations.
6. Label to Questions: help a parent understand a medicine leaflet
Your parent has a leaflet full of unfamiliar words and a short appointment coming up.
Retrieve the official drug label through Valyu, then use a local open model to explain selected sections and turn confusing terms into questions for the pharmacist. The app should preserve the original text beside the explanation.
Support one medicine and three sections: warnings, interactions, and administration information. Demonstrate that a missing answer remains missing. The app helps someone understand and ask; it does not change their prescription.
7. Grant Scout: funding research for a friend's community project
Your friend runs a small community organisation and cannot spend every evening hunting for grants.
Valyu Search finds current programmes; Contents reads funder pages for geography, eligibility, award size, and deadlines. An open model compares the published requirements with the organisation's work and drafts a gap checklist.
Start with one cause and region, and return five opportunities. Show why one funder fits and another does not. The deliverable is a shortlist your friend can act on, not an invented application or a claim that funding is guaranteed.
8. Prior Art Pal: a research notebook for your inventor friend
Your friend is building a hardware prototype and wants to know what already exists.
Use Valyu's patent and academic coverage to find similar mechanisms, then let an open model compare them with a plain-language description of the prototype. Preserve publication numbers, dates, figures where available, and source links.
Ship a notebook of five relevant references and an editable feature comparison. Demo the difference between "similar idea" and "same mechanism". This is research support, not a patentability or freedom-to-operate determination.
9. Before You Quit Your Job: a startup pre-mortem
Your friend has a business idea, a spreadsheet, and dangerously enthusiastic group-chat support.
Use Valyu DeepResearch to investigate comparable businesses, documented failures, and market constraints. An open model turns the report into an interactive pre-mortem: which assumptions are unsupported, which warning signs matter, and what small experiment could test each assumption?
Build around one idea, five comparable cases, and three experiments. Demo a question your friend had not considered, backed by a real case. Global Fail Map is a useful reference for this evidence-first approach.
10. Neighbourhood Numbers: a competitor notebook for a café owner
Your friend wants to open a café and has researched the neighbourhood mainly by drinking in it.
Use Valyu to discover nearby competitors and extract published menus, opening hours, and positioning. An open model organises a cited comparison and identifies information that still needs a visit or phone call.
Start with one neighbourhood and five businesses. Show the source and check date for every price. If you already have a suitable historical sales dataset, TabPFN could add a separate forecasting experiment; do not turn five menu pages into a pretend demand model.
11. Repair Rights: a source-linked checklist for your renting friend
Your friend has a broken boiler and no idea which message to send their landlord.
Use Valyu Search and Contents to retrieve official housing guidance for one jurisdiction. An open model turns that guidance into a dated checklist and a draft message your friend can edit, with sources beside the relevant steps.
Ship one repair scenario in one city or country. Demo a jurisdiction mismatch being rejected. This should help someone organise information and seek appropriate help, rather than manufacture legal certainty.
12. Landing Pad: a relocation brief for a friend moving abroad
Your friend is moving for a course or job and keeps asking the same questions in different group chats.
Use Valyu DeepResearch to investigate official arrival requirements, university or employer guidance, housing resources, and transport options. An open model converts the findings into a personalised arrival checklist and quizzes your friend on what is still unresolved.
Focus on one destination and one arrival date. Demo an expired requirement being replaced with current official guidance. Keep required documents separate from optional suggestions, and flag anything that needs confirmation.
13. Access Before Arrival: travel planning for a wheelchair-using friend
Your friend needs to know whether a venue will work for them before paying for the trip.
Use Valyu Search and Contents to gather venue access statements, transport guidance, and relevant visitor information. An open model compares those statements with your friend's requirements and prepares specific questions for the venue.
Research three venues in one city. Demo a useful distinction between "wheelchair accessible" and documented details about entrances, toilets, or lifts. Unknown access stays unknown until someone verifies it; the app must not invent step-free routes.
14. Our Street Has a Story: a local-history walk for your grandparent
Your grandparent loves a neighbourhood, but the interesting history is scattered across archives and museum pages.
Use Valyu DeepResearch to investigate five nearby landmarks. An open model turns the cited reports into short walking-tour stories at a reading level or in a language your grandparent chooses.
Ship five stops with sources and an optional narration layer. Demo the original evidence beside the shorter story. History offers a useful map-to-research pattern, but your personal route and open-model storytelling are the new project.
15. Recipe Roots: a sourced family cookbook
A relative has recipes in voice notes, with instructions like "put enough until it looks right".
Use an open speech-to-text model to transcribe the recordings locally. Valyu retrieves regional culinary references and technique explanations; an open language model organises the recipe and separates your relative's words from researched context.
Start with three recipes and an editable export. Demo a family instruction that remains unchanged beside an attributed explanation. Missing quantities should become questions for the cook, not measurements the model confidently makes up.
16. Pitch With Receipts: a research pack for a journalist friend
Your friend has a promising story idea but needs to know which parts survive checking.
Use Valyu to find primary documents, relevant studies, and reporting on the subject. An open model creates a claim ledger: the proposed claim, supporting source, conflicting evidence, and unanswered questions.
Limit the build to one pitch and ten claims. Demo one claim being weakened after contradictory evidence appears. DeepResearch can prepare the background dossier; the open model powers follow-up questions and keeps the ledger useful during editing.
17. Is This Offer Real?: a scam-evidence notebook
Your friend receives a job or investment offer and wants to investigate before responding.
Use Valyu Search and Contents to retrieve official company pages, regulator warnings, and other verifiable records. An open model organises inconsistencies, unanswered questions, and the safest next verification steps without sending your friend's private messages into search.
Ship a URL-based investigation with source-linked findings. Demo a mismatch between the offered contact details and the official site. Missing evidence does not establish legitimacy, and suspicious evidence is not a licence to accuse someone without verification.
18. Recall Radar: a used-car research helper
Your friend has found a used car and wants more than the seller's "drives beautifully".
Use Valyu DeepResearch to investigate the model year through NHTSA recall, complaint, and investigation records, alongside manufacturer information. An open model produces an inspection question list and explains the difference between a complaint, a recall, and a confirmed repair.
Start with one model year. Demo a recall linked to the official record and a reminder to confirm applicability and completion for the actual vehicle. Automotive datasets are DeepResearch-only in Valyu's current catalogue.
19. Patch Me First: a dependency-risk brief for your developer friend
Your friend maintains a small app and cannot investigate every vulnerability notice in depth.
Read their dependency list locally. Use Valyu to retrieve relevant advisories and official remediation guidance, then have an open model explain version applicability, documented mitigations, and unresolved questions.
Limit the first version to five dependencies and a manually reviewed patch queue. Demo an advisory that does not apply to the installed version. The useful output is a cited explanation, with code-based version checks, rather than a model guessing which package looks scary.
20. Don't Kill the Tomatoes: an evidence-backed garden helper
Your friend has a balcony garden and is about to follow a confident, contradictory internet diagnosis.
Use Valyu Search to retrieve crop-specific guidance from agricultural extension services and research sources. An open model asks symptom questions, compares possible causes, and explains what observation would help distinguish them.
Build for one crop in one region. Demo two plausible causes with different evidence and an explicit uncertainty label. If you already own an Arduino UNO Q, add real soil or light readings and a meaningful on-board model; that could support the Arduino category.
Which one should you pick?
Choose the idea for which you can reach a real user today and verify the underlying sources yourself.
For a relatively small first build, I'd start with Interview Espresso, Grant Scout, a local-history walk, or a tightly scoped paper queue. Trial research and patent comparisons demand more domain review. Hardware only makes sense if the hardware is already on your desk.
Ask your friend one question before you open your editor:
"Show me the last time this was annoying."
You'll learn more from watching that than from asking whether they would use an AI app.
You do not have to invent every pattern: 11 apps with public code to explore
I've built research tools on my own and with my friend yorkeccak, and we've made the code available so other developers can inspect and build on it. The Valyu example-app gallery lists ten apps; Global Fail Map belongs on your reading list too.
For a closer look at the product decisions, I've written about building Global Fail Map and DeepResearch for life sciences in Bio.
Here is the full set, with two sentences on what each does and links to both the app and codebase.
One reuse detail matters: Bio, History, Finance, PatentAI, and Global Fail Map have MIT licence files in their repositories. The other six have public code, but their licensing documentation is incomplete or inconsistent: some READMEs say MIT without the referenced licence file, while Global Threat Map's README says MIT and its package metadata says ISC. Confirm the licence with the maintainers before reusing those projects; public visibility alone does not establish an open-source licence.
Polyseer
Polyseer takes a Polymarket or Kalshi market URL and uses Valyu DeepResearch to investigate the evidence behind the outcome. It returns a sourced forecast, supporting and opposing factors, and downloadable research files, making it a useful example of turning a specific external object into a research task.
- App: polyseer.xyz
- Code: github.com/yorkeccak/polyseer
Bio
Bio runs biomedical research workflows across sources such as PubMed, clinical-trial records, drug labels, patents, and the web. It returns cited reports and deliverables for tasks such as trial mapping, pipeline analysis, and regulatory research, with a workflow browser and research activity feed.
- App: bio.valyu.ai
- Code: github.com/yorkeccak/bio
History
History lets you select a location on an interactive globe and investigate its history through Valyu DeepResearch. It connects geographical exploration with cited reports and saved discoveries, which is a useful pattern for a personal history tour or classroom project.
- App: history.valyu.ai
- Code: github.com/yorkeccak/history
Finance
Finance's current implementation runs Valyu DeepResearch from a research question or a financial workflow, with source-linked reports and a live activity feed. It provides report history, downloadable deliverables, and example reports for areas such as investment banking and private equity, making it a useful reference for a financial research workflow.
- App: finance.valyu.ai
- Code: github.com/yorkeccak/finance
Patents / PatentAI
PatentAI lets users search and compare USPTO and EPO patent material in natural language, including patent drawings where available. It combines Valyu retrieval with analytical tools and local-model options, offering patterns for prior-art notebooks and source-linked technology comparisons.
- App: patents.valyu.ai
- Code: github.com/yorkeccak/patents
Global Threat Map
Global Threat Map plots reported security events and geopolitical developments on an interactive map. It combines Valyu Search, Answer, and DeepResearch for event discovery, country context, and sourced intelligence dossiers with exportable files.
Competitor Analysis
Competitor Analysis turns a competitor's website and research context into a sourced report about its products, positioning, and recent developments. It demonstrates an asynchronous DeepResearch workflow with progress updates, a readable report view, and PDF output.
Supplement Research
Supplement Research is designed to investigate scientific evidence, safety information, and published product details for a supplement. Its DeepResearch integration requests a report alongside a brand-comparison CSV and a short document, showing how one investigation can request different useful outputs.
- Listed hosted URL: supplement-research.up.railway.app
- Code: github.com/unicodeveloper/supplement-research
Meeting Intel / Intel Espresso
Meeting Intel, named Intel Espresso in its repository, creates a short briefing on a company or topic before a meeting. It uses Valyu's Answer API to organise recent developments, relevant people, dates, talking points, and source links into a print-friendly brief.
- Listed hosted URL: meeting-prep.up.railway.app
- Code: github.com/unicodeveloper/meeting-prep
Consult Ralph
Consult Ralph runs DeepResearch for consulting and analytical tasks such as company due diligence, market analysis, and competitive research. It accepts research focus, client context, and source attachments, and requests reports plus CSV tables, executive-summary documents, and presentation files.
Global Fail Map
Global Fail Map is an interactive atlas of failed companies, cancelled megaprojects, abandoned technologies, science programmes, and unrealised visions, with 200 curated stories in its bundled collection. It pairs cited case reports with live Valyu DeepResearch investigations into what was attempted, why it ended, what survived, and what a future builder can learn.
How to make a strong case on every judging criterion
No article can guarantee a prize. What you can do is make your entry easy to understand and give the judges evidence for each criterion.
1. Writing quality: tell the story of the person, then explain the system
This carries the most weight. Budget time for it before you add another feature.
Compare these openings:
"I built an innovative research application using modern AI technologies."
And this illustrative opening:
"My sister found a scholarship she wanted, then discovered the deadline had passed. I built her a shortlist that checks official requirements and makes the next deadline hard to miss."
The second gives me a person, a problem, and a reason to care. Use an incident that actually happened to your friend, rather than copying the example as your own experience.
Then answer the obvious reader questions: What does the app do? What happens after I click? Where did the information come from? What did you change after your friend tried it?
Put a short demo near the top. Include an architecture sketch(excalidraw is a great example for it), a useful code excerpt, one bug you fixed, and one limitation you still have. Record genuine feedback; do not invent a touching quote because the deadline is close.
A good title names the person or task: "I built my dad a medicine-leaflet explainer that shows its sources" tells me more than "Introducing HealthMate 3000".
2. Relevance to the prompt and theme: prove both halves
Your entry needs a real friend problem and a meaningful open AI core.
Show the actual task your friend performs. Explain the model or framework, where it runs, and which functionality depends on it.
Then explain why openness matters for this person. Maybe you can swap models when their laptop struggles, keep their profile local, inspect the workflow, or tune the reading level. Demonstrate the benefit you claim. "It runs offline" is not accurate if every question still calls a hosted search service.
3. Creativity: let your friend's constraints shape the interface
You do not need to invent a new category of software. You need a fresh, useful response to a real situation.
One friend needs a bilingual arrival checklist. Another wants the original drug-label wording visible beside the explanation. Your grandparent may prefer a printed walk with large type to a chat window.
That specificity is where the interesting design comes from.
Show the detail that a generic app would miss: an "unknown accessibility" state, a jurisdiction check, a contradiction view between papers, or a recipe quantity that becomes a question for the family cook. Creativity is easier to judge when it changes what the user can do.
4. Technical execution: make the boring parts trustworthy
A good demo completes one useful task without hidden manual rescue.
For these projects, I would prioritise working source links, structured fields, accurate dates, clear errors, and saved research tasks. Handle an empty result and a failed request as deliberately as the happy path.
Use code to enforce constraints such as valid citation IDs, supported jurisdictions, version ranges, and required source URLs. Models can explain the evidence; they should not be the only thing checking a deadline or doing arithmetic.
Test on your friend's device. Record real latency and cost if you include those numbers. Show a failure case in the write-up and explain what your app does about it.
For DeepResearch, a reload should reopen the existing task rather than launch another paid investigation. That small implementation detail is more convincing than a screenshot with an invented "99% complete" bar.
5. Use of partner technology: show the contribution, not the logo
This criterion applies when you enter a partner category. Valyu is not listed among this challenge's prize partners, so a Valyu integration does not create a "Best Use of Valyu" category.
You can combine Valyu with a listed partner where the fit is real. For example:
| Official category | Meaningful use in a Valyu-powered project | Evidence to put in your post |
|---|---|---|
| Gemma | Runs your personalised explanation, interview rehearsal, or research dialogue | Model/version, code path, and an actual output |
| Render | Hosts the app, agent runtime, or front end | Deployment details and a working demo |
| DigitalOcean | Hosts your app or serves an open model | Running architecture and deployment evidence |
| Mastra | Orchestrates an open-model workflow with retrieval and follow-up tools | A workflow trace and the relevant implementation |
| Temporal | Keeps a research workflow recoverable across failures | A demonstrated interrupted-and-resumed run |
| MongoDB Atlas | Stores research history, retrieval data, or agent memory | Schema and a working retrieval or history feature |
| Tiger Data | Supports embeddings or hybrid retrieval over saved research | The query and the result it enables |
| Sentry Agent Tracing | Reveals latency, tool calls, and a problem you debugged | Real traces or screenshots and the resulting fix |
| ElevenLabs | Adds narration or a useful voice interface | Audio and an explanation of its role |
| TabPFN | Predicts from an appropriate historical tabular dataset | Dataset, baseline, evaluation, and limitations |
| Tinker | Fine-tunes a model for a specific task | A fair before/after comparison on held-out examples |
| Arduino | Uses an UNO Q for a meaningful physical agent or on-board model | Hardware demo and actual sensor/model behaviour |
The remaining categories and their exact requirements are on the official hub. If you enter SerpApi alongside Valyu, for example, give it a distinct, implemented search role and show what it contributes. Installing a package is not the same as using it meaningfully.
Pick integrations because they improve the project. Twelve credentials and an unfinished demo make a miserable Sunday.
A weekend plan that leaves time for the article
Treat the following as a suggested schedule, not a requirement:
| Time | What to finish |
|---|---|
| First hour | Watch your friend do the task; define one measurable success condition |
| Next two hours | Retrieve real evidence with Valyu and run the open model on it |
| Rest of Saturday | Build one complete user flow, including source links and an empty/error state |
| Sunday morning | Put it in your friend's hands; observe and fix the biggest problem |
| Sunday afternoon | Record the demo, document setup, and write the submission |
| Before the deadline | Check the published post, repository, demo access, tags, and claimed prize categories |
A suggested success condition for Deadline Buddy could be: "My sister can find an eligible opportunity and its official deadline without opening ten tabs." Measure what happened during her test; do not announce a time saving you did not observe.
Your feature list can wait. Your friend and the judges need one complete outcome.
What your submission should contain
Use the submission-template button on the challenge page. Its sections are:
- What I Built: name the friend or loved one, their problem, and the app's useful outcome.
- Demo: include a deployed link or video showing the actual workflow.
- Code: link the new repository and explain how to run it.
- How I Built It: explain the open model or framework, Valyu's role, and the architecture.
- Why Does Open Innovation Matter?: describe a concrete benefit and show evidence for it.
- My Agent Session: optionally link or embed your build session using the challenge's instructions.
- Prize Categories: list only the categories your implemented project genuinely uses.
The template adds devchallenge, weekendchallenge, and hf26challenge. Add those to your actual challenge submission. This idea article is a guide, not a completed challenge entry.
Build something they ask to keep
I've built maps of threats, maps of failure, research tools, and brief generators. The part I care about is what someone can do with the result.
A cited report is useful when it helps a friend ask a better question. A spreadsheet is useful when they can compare the options. A rehearsal is useful when they feel prepared enough to close the laptop.
Pick your person. Watch the annoying task. Give your open model real evidence through Valyu. Then hand the project over and see what happens.
If your friend asks, "Can I keep using this?", you have a very good ending for your article. And a reason to maintain the project after Monday.
Top comments (0)