I’ve always liked taking the time to plan things out before jumping into a project. Lately, I’ve been thinking about what would make a great first project to try out what I’ve learned about Azure’s AI tools—and something that would also be a solid addition to my portfolio.
As commented on my last post, I would like the project to respect the following guidelines:
- It should be low cost : For obvious reasons. (Reliving the student budget approach here).
- Time efficient: While I do want it to challenge me and expand my comfort zone, I want to keep the timeline for this project as short as realistically possible. My action plan contains many items to be tackled and my available time, after work and parenthood is…definitively, limited.
- No doorkeepers: As you may have read on my first post, I have a sales background. I know what it takes to sell an idea. I don’t have the time for all the needed prospecting and then presenting a project to a company, following up and having to go through their decision making and approval processes. That would take a huge amount of time and effort. Multiple meetings with stakeholders, and eventually, it will be extremely sensitive to budgets. Even proposing it as labor-free, it will imply some kind of a CapEx from the client.
My plan is to keep this one private, pay for my own development expenses (i.e., paid AI services, hosting, etc), document everything and then kill all the paid services when I finish. - Trained with publicly available data: To make the first bullet point possible I need to use easy-to-access-publicly-available data instead of costly REST API calls to pull data from other sources.
- There should be a Strong Justification to invest in its development. I think there’s not much logic in developing an app that does something that can be asked to ChatGPT, Gemini or any of the other AI chat bots directly.
Thoughts on the limitations of generic AI chat bots
- General-purpose chatbots are broad but not specialized.
- Can’t monitor data feeds continuously for you.
- Lacks fact retrieval natively.
- Sourcing facts, showing evidence, highlighting comparisons.
- Not easy to paste multi page documents to a chat prompt.
- UI limitations: Not easy/practical/economical to upload multiple documents to the chat for analysis.
- Not integrated with 3rd party software like CRMs, Dashboards, notifications, etc.
These are the first ideas that came to mind
1.- F.A.Q. chat-bot for customer support
Problem
“Corporate F.A.Q. pages are static, often incomplete, and able to help only the users that speak the language they were drafted on. Most FAQ pages are really not that useful, defeating their purpose of supporting the customer base and decreasing the load of calls to one on one customer support calls ”
Concept
- Trained with the content of their current available FAQ page, brochures, support emails and documents, this chatbot would be exponentially more thorough than a static FAQ page on a corporate website. It would be cable to support users from multiple languages without having to translate the content.
- Possible module filtering communications from human customer service agents.
Justification score: High
Feasibility score: MEDIUM
Why the score?
Pros
- A user would get a proper response quicker and more accurately in a pre-trained chat-bot than on a generic AI chat-bot. As the generic chatbot would need to do some due diligence and web scrapping to get all the facts.
- Fairly easy to build if customer support questions and answers are provided by the customer.
- Data is usually available and free.
- A customer support AI bot will be able to answer requests in any language, not only the one used to create the material.
- The LLM to be used will allow the bot to “understand” semantics of the questions and even paraphrase responses if the user is not understanding.
Cons
- Since garbage in is garbage out, the quality of the data to be used to feed the model is critical. Obtaining the right customer support data will imply a serious collaboration with the customer support and possibly tech support teams of the client company.
2.- Bi-laws and regulations Genie
Problem
“In most cities, there isn’t a single, centralized place where you can check whether something is legal under local regulations. Instead, the information is often scattered across multiple websites, requiring indexed searches that lead to different sources—each offering only part of the answer”.
Concept
- A chat bot trained with all the bylaws and regulations of the city were I live (Kitchener, Ontario, CA)
- The Language model will cut user’s research time in less than half, and more than double the thoroughness of the information. Even further, it will decode the semantics of the user’s prompt and provide the proper answer based on the data it was trained.
- Multiple language support thanks to a translation model.
Justification score : HIGH
Feasibility score : LOW
Why the score?
Pros
- Pretty easy to build if data available.
- Data is usually available and free.
Cons
- Possible interference from the city?
- Data QualityConfusing / Inconsistent Data?
Facts
- Any of the chats would be too generic and would take too much time to build a proper prompt to get the answers that a pre-trained chat-bot would generate if pre-trained with all the data.
3.- Actor Screen Time Monitor
Problem
Screen time is a very important metric used by studios, TV, and movie industry companies to measure character prominence, identify imbalances in representation, (i.e., whether certain gender groups or ethnicities are underrepresented), and reveal how writers and directors prioritize specific characters and storylines (Narrative).
This is Often manually tracked by watching the media and recording the duration of each character’s appearances. Manually tracking screen time can be time-consuming and prone to inaccuracies. “
Concept
- This app would use visual recognition models to analyze an entire movie or TV episode, identifying the actors in each scene and keep track of how often each actor appears on screen, for how long, whether they are alone or with others, and which other actors they share scenes with.
Justification score: MEDIUM
Feasibility score: LOW
Why the score?
- Pros
- More practical and time efficient to use a trained-specific app than to do in any of the existing AI Chats.
- Chat GPT reported several limitations like:
- Not being able to calculate screen dominance (.g., how prominent the character is in a scene)
- Only capable of analyzing clips of under 2 mins
- The analysis won’t be frame-perfect.
- Not sure if able to process costume changes accurately
- Not many competing apps out there doing this
- Cons
- Very expensive in computing power and services.
- Probably very expensive in development costs and AI services
4.- Corporate “Kosherizer”
Problem
” All companies—regardless of industry—care deeply about how their representatives communicate with the outside world. These communications must align with the company’s values, technical guidelines, and often legal requirements.
As a result, in more conservative organizations, even minor communications or marketing materials typically require review and approval by the legal department to proactively avoid risk and liability.
While this level of caution is understandable, it often creates a bottleneck. The approval process can introduce significant delays, making it difficult to get timely marketing or sales content out the door. As a result, employees tend to fall back on outdated collateral, slogans, or messaging simply because they’ve already been approved.
For example, a salesperson might send an email to a prospective customer using old materials that describe features no longer offered—just because it’s safer and faster than waiting for updated content to be reviewed
On the flip side, the need for a strong review process in company communications becomes clear when you consider the risks of unfiltered messaging. Imagine a new salesperson posting a product description on LinkedIn that significantly overstates the product’s capabilities—ignoring its real limitations. This kind of miscommunication can damage the company’s credibility, create false expectations, and even lead to legal or contractual issues.”
Concept
- Company lingo, values, technical, and legal moderator.
- Review company communications (emails, letters, collateral) proofing the syntax, meaning, and message to make sure it matches company messaging, values, limitations and product specs. (company-specific).
- Able to analyze multiple format of documents, images, etc.
- Trained with company pre-approved emails, collateral, mission statement and more.
Justification score: High
Feasibility score: Low
Why the score?
Pros
- Developing an app is justified.
- While possible, it is not practical to bring up to speed a chat bot on the specifics of a company to perform this task.
- The app’s UI would facilitate the upload of documents.
Cons
- Need to collaborate with multiple departments.
- Liability concerns if the app greenlights a communication with a subtle but meaningful semantic implication that the AI didn’t catch.
Facts
- It would require some type of human final approval to avoid the risks of “AI hallucinations” affecting the result.
5.- Due Diligence Genie for Stock Traders
Problem
” Intra-day stock traders—especially short sellers—have very limited time to perform proper due diligence before entering a trade. From the moment an alert pops up on their scanners to the opening bell, they need to quickly determine whether the company behind the ticker meets all of their criteria. Time is tight, and there’s a lot to get done—like locating borrows to short, setting up orders, and refining their strategy.
Depending on the trader’s approach, the key parameters they might need to assess include: country of issue, float size, catalysts, recent news, market cap, short interest, institutional and insider ownership, operating cash flow, potential dilutions, cash reserves, accumulated deficit, sector trends, and more. All of this has to be evaluated on the fly—while simultaneously applying the rules of their trading strategy to decide whether to enter the trade, how much to size in, and how to manage risk.”
Concept
- Customer will input the ticker or set of tickers they are considering to trade
- Rule-based part of the app will pull all the needed info from different sources via API and create a data set
- Dataset will be run by a model pre-trained with all the strategy’s rules and considerations and it will generate a report highlighting where the data is compatible with the trading system and what items represent a violation or red flag to consider.
Justification score: HIGH
Feasibility score: HIGH
Why the score?
Pros
- My wife is a short-biased intra-day stock trader. that gives me full access to her strategy, rules, and input.
- This is something that can’t be done by a generic bot due to all the API’s and strategy rules involved.
- This will help my wife make money. (This might end up helping my case of buying a boat)
Cons
- Very niche specific
- Very specialized
Then, it dawned on me…
What if I asked all the major AI bots for project ideas, including my own guidelines in the prompt?
Wouldn’t that be exactly the kind of thing an AI hero would do?
This is what Chat GPT came up with
1.-PDF report comparator
Details
Problem
Detecting changes in large legal documents is a complex and time-consuming task. It’s crucial to ensure that no modifications have been made before signing or sharing a document externally. Manually comparing documents significantly slows down the countersignature process.
Concept
- Upload two PDFs (e.g., annual reports, policies) and see what’s changed — additions, removals, and rewritten sections.
Justification score : LOW
Feasibility score : LOW
Why the score?
Pros
- There’s a strong case for using AI to perform this task.
Cons
- I believe it’s something ChatGPT, Gemini, or Copilot can handle easily. You can upload—or even paste—both documents and have the model analyze the differences
Facts
- To justify developing this app, it must offer a clear advantage to the user. Would it be faster than simply uploading two documents to ChatGPT? Would the results be clearer, perhaps by using a custom-trained model tailored for this specific task?
2.- Health Misinformation Detector
Details
Problem
False or misleading health claims spread rapidly online, leading to real-world harm like delayed treatment or vaccine hesitancy. Manual fact-checking can’t keep up with the volume of content. This project aims to build an AI-powered tool that flags potentially misleading health information and cross-checks it against trusted public sources (like WHO or CDC), helping users quickly identify and understand questionable claims.
Concept
- Critical in health communication.
- Checking claims against trusted sources, not just generating a “maybe” answer.
- Adds fact retrieval that ChatGPT lacks natively.
- Checking claims against trusted sources, not just generating a “maybe” answer.
Justification score: LOW
Feasibility score: LOW
Why the score?
Pros
- Adds fact retrieval that Chat GPT lacks natively.
- Using technology (AI) to solve the misinformation pandemic generated by the wrong use of technology (Internet).
Cons
- For the same reason people struggle to fact-check questionable claims, feeding the model with the right data might be complex and time consuming.
- Not to sound conspiratorial, but this topic is known to generate enemies.
3.- Local Government Transparency Dashboard
Details
Problem
Local governments publish critical information—such as budgets, contracts, and council meeting minutes—but this data is often buried in PDFs or outdated websites, making it hard for citizens to access, search, or understand. The lack of accessibility and clarity creates a transparency gap, limiting public oversight and engagement
Concept
- Automatic summarization of civic meeting minutes into trends and alerts.
Justification score: LOW
Feasibility score: LOW
Why the score?
Pros
- Developing an APP with its own UI is justified. No one is going to copy-paste 50 PDFs into ChatGPT manually.
Cons
- Not sure I want to pick a fight with the local governent.
4.- Contract Clause identifier
Details
Problem
Contracts are often long, complex, and filled with dense legal language, making it difficult and time-consuming for non-lawyers—and even busy professionals—to locate and understand specific clauses related to risk, liability, confidentiality, or termination. Manual review is error-prone and inefficient, especially when comparing multiple documents
Concept
- Fast legal clause extraction + explanations
- Real-world document uploads.
Justification score: MEDIUM
Feasibility score: MEDIUM
Why the score?
Pros
- There’s for sure a place in the Canadian market for an app like this one trained and validated with Canadian law. Most apps were developed with the US market in mind.
Cons
- Very Similar to # 1,
- Complex to validate the data to be used to feed the model. Probably would need to hire/collaborate with a lawyer.
- I repeat: collaborate with a LAWYER!
Facts
- Chat GPT doesn’t easily compare two contracts or highlight differences.
- Unless sure of the sources. a generic chat bot responses can be a hit and miss.
5.- Public tenders analyzer
Details
Problem
Public tenders contain valuable information about government projects and spending, but they are often published in unstructured formats (PDFs, scanned docs) and spread across different platforms. This makes it difficult for businesses, watchdogs, and researchers to efficiently identify relevant opportunities, track trends, or detect irregularities. Manual analysis is slow and inefficient.
Concept
- Analyze open government tender documents to extract the category, budget, timeline, and other details.
- Structure hundreds of messy government PDFs automatically.
- Companies need this for biz dev .
Justification score: MEDIUM
Feasibility score: MEDIUM
Why the score?
Pros
- Government procurement data is usually public.
- A custom UI would be justified, using Chat GPT would be too manual and building the right prompt can be complex.
Cons
- I have the feeling that there has to be many solutions for this already in the market
6.- Resume Analyzer & Job Matcher
Details
Concept
- Specialized skill extraction + matching against specific job datasets.
- Personalized career help.
Justification score: LOW
Feasibility score: HIGH
Why the score?
Pros
- Way more focused than general advice from ChatGPT.
Cons
- Crowded space in AI applications
- Multiple players started early building their apps.
7.- Sales Intelligence Aggregator
Details
Concept
- Scrapes and summarizes competitor news, public earnings reports, customer reviews for a targeted company list.
- Full automation: You could have the app to scrape news, reviews, earnings data on a schedule
- Custom filters (e.g., only summarize negative reviews, or compare sentiment YoY)
- Persistent memory of companies, context, benchmarks
Justification score: MEDIUM
Feasibility score: LOW
Why the score?
Pros
- Development of an independent app is justified as a chat-bot on its generic state can’t do this task properly
- Real-time scrapping only possible trough an app.
- Highly customizable
- A dedicated model trough an app can be re-trained with responses so it hold information.
- Scalable across many companies
Cons
- Extensive market research needed.
- Very crowded space by big players (LinkedIN, ZommInfo, etc.).
- Requires some serious development time.
- Requires participation and vetting from the client’s sales team.
- Slight infrastructure cost (scraper, database, GPT API).
- You may need to handle rate limits, data cleaning, summarization rules, etc.
Facts
- ChatGPT can’t actively collect and organize external real-time data across companies without external scripting.
8.- Custom Compliance Checker
Details
Problem
Organizations struggle to consistently ensure that internal documents, contracts, and workflows comply with complex and evolving regulatory requirements. Manual compliance reviews are time-consuming, error-prone, and often require legal expertise, leading to increased risk of fines, legal exposure, and reputational damage.
Concept
- Automatically reviews documents or workflows and flags compliance risks.
Justification score: HIGH
Feasibility score: LOW
Why the score?
Pros
- App can continuously monitor or batch-process large volumes of documents 24/7 without user intervention.
- App: Can enforce custom compliance rules, thresholds, industry standards, and internal policies, tailored to your business.
- App: Can integrate with document management systems (SharePoint, Dropbox, GDrive), CRMs, ERPs, etc.
- App: Can log results, generate PDF reports, dashboards, or notifications for flagged risks
- App: Can serve multiple users with access controls, roles, and versioning.
Cons
- Cost if looking to scale: Running certain LLMs at scale can be costly
- Legal & Regulatory Responsibility: If your app flags (or fails to flag) a compliance issue, who is liable? Especially in industries like healthcare or finance
- Disclaimers, legal review, or even liability insurance might be needed
- Accuracy & False Positives/Negatives: LLMs can miss nuanced legal phrasing, or over-flag safe content
- The model could recommend users to do unnecessary work or, worse, missed real issues.
- You’ll likely need a human-in-the-loop (HITL) system or some post-processing logic.
- Competition: There are already players in this space (e.g., Evisort, Legalsifter, Compliance.ai, and others).
- User Trust & Adoption: Legal teams or compliance officers may distrust AI-only systems
Facts
- ChatGPT can’t consistently enforce compliance rules or track/check full documents or datasets.
- Requires manual uploads and instructions per session
- Chats need to be told what to check every time
- Chats don’t integrate with other platforms
- Chats are single user interfaces
9.- Enterprise Meeting Summarizer + Action taker
Details
Concept
- Parses Zoom/Teams meeting transcripts into clear summaries, action items, and owner assignments.
Justification score: LOW
Feasibility score: LOW
Why the score?
- Pros
- Development of an independent APP would be justified. ChatGPT can summarize text but won’t track who owns tasks across multiple meetings.
- Cons
- Crowded space: I know of at least a couple product in the market for this. A differentiation could be to make it company-specific
- Expensive in APIs and services
This was Gemini’s brainstorm
1.- Automated Customer Feedback Analysis (Sentiment & Topic Dashboard)
Details
Problem
“We get customer feedback from reviews, surveys, support tickets, and social media, but it’s overwhelming to process manually. We struggle to quickly understand sentiment, identify recurring issues, or spot emerging trends.”
Concept
- Stop drowning in feedback and start acting on insights.
- This AI solution automatically analyzes incoming customer feedback (from various sources) to determine sentiment (positive, negative, neutral) and extract key topics and phrases. Get a clear, real-time view of customer opinions, identify urgent issues faster, and track trends to inform product development and customer service strategies
Justification Score: MEDIUM
Feasibility Score:
Why the score?
- Pros:
- Companies manage mountains of documents (reports, contracts, research, manuals, knowledge base articles)
- Finding specific information is like finding a needle in a haystack.
- Teams waste hours searching, and valuable insights remain buried.”
- Cons:
2.- Image & Digital Asset Analysis
Details
Problem
“We have a large library of images (marketing photos, product shots, historical scans) but it’s hard to manage and search effectively. We need a way to automatically tag them, find specific content, or even ensure compliance (e.g., check for inappropriate content or specific logos).”
Concept
- Bring order and intelligence to your digital asset management.
- This solution uses AI to ‘see’ your images, automatically generating descriptive tags, detecting objects and text, and enabling content-based search.
- Find the right image instantly, streamline workflows, and ensure brand consistency or compliance.
Justification Score: MEDIUM
Feasibility Score: LOW
Why the score?
- Pros:
- Very relevant for marketing departments, e-commerce businesses, media companies, archives, and potentially manufacturing (inspecting product images)
- Very relevant for marketing departments, e-commerce businesses, media companies, archives, and potentially manufacturing (inspecting product images)
- Cons:
- Potential for Inaccurate or Embarrassing Tags: AI can make mistakes
- Expensive to develop (Initial Processing Costs) and to maintain (Ongoing operational costs, Data storage, etc.)
- Client might be worried about data privacy concerns (sensitive images).
Discarded Ideas similar to others already considered
- Intelligent Document Search & Analysis
- Sentiment Analysis Dashboard for Public Feeds
- 24/7 Intelligent FAQ Bot (Internal or External)
- Automated Data Extraction from Documents
This was Grok’s grain of sand
1.- Social Media Trend Analyzer
Details
Problem
“Businesses need to monitor social media trends to understand public sentiment and emerging topics, but manually analyzing large datasets like tweets or Reddit posts is time-intensive, and real-time API access to platforms is often costly or restricted”
Concept
- Develop a Social Media Trend Analyzer using Azure Text Analytics to extract key phrases, sentiment, and entities from a publicly available dataset, such as the Twitter Sentiment140 dataset from Kaggle.
- Use Azure Functions for serverless data processing, store results in Azure Blob Storage, and create an interactive dashboard with Power BI (free tier) to visualize trending topics, sentiment distribution, and entity frequencies.
- Host a static documentation site on Azure Static Web Apps (free tier) to showcase the project. All services will be terminated after documentation to avoid costs.
Justification Score: LOW
Feasibility Score: HIGH
Why the score?
- Pros:
- Strong Justification: Identifying trends in social media data is a high-value business use case, showcasing Azure’s NLP capabilities beyond general-purpose chatbots.
- Low Cost: Uses Azure’s free tier (Text Analytics, Functions, Blob Storage, Static Web Apps) with costs under $5 if within limits.
- Time Efficient: Can be completed in 12-15 hours over 2 weeks, including data processing, dashboard creation, and documentation.
- No Doorkeepers: Self-contained, requiring no external approvals or client interactions.
- Public Data: Relies on freely available datasets, avoiding paid API calls.
- Cons:
- Potential for Inaccurate or Embarrassing Tags: AI can make mistakes
- Expensive to develop (Initial Processing Costs) and to maintain (Ongoing operational costs, Data storage, etc.)
- Client might be worried about data privacy concerns (sensitive images).
Claude’s contribution
1.- Computer Vision Quality Control System
Details
Problem
“Manual visual inspection in manufacturing is inconsistent, slow, and prone to human error, especially for high-volume or precision products.”
Concept
- An automated visual inspection system using Azure Custom Vision, Azure IoT Edge, and Azure Cosmos DB.
- Deploys trained models to edge devices on production lines.
- Detects defects in real-time with greater accuracy than human inspectors.
- Aggregates quality metrics and identifies process drift.
- Continuously improves through feedback loops.
Justification Score: HIGH
Feasibility Score: LOW
Why the score?
- Pros:
- Reduces defect escape rates.
- Lowers inspection costs.
- Provides data-driven insights for process improvement.
- Allows 100% inspection coverage vs. random sampling
- Cons:
- Expensive Data
- Requires large datasets of defective and non-defective products for training
- Collecting enough data can be time consuming
- Labeling defects accurately requires domain expertise and is labor-intensive
- Small Changes can afftect results
- Lighting variations
- Temperature and humidity
- Different product batches, materials, or colors
- Dust, vibrations, or camera positioning
- High Initial Investment
- Specialized cameras, lighting systems, and edge computing hardware are expensive
- Integration with existing production lines requires significant engineering effort
- Training and deployment can take months, delaying ROI
- May require production line modifications or temporary shutdowns
- Expensive Data
Discarded Ideas due to complexity or because they were similar to others already considered
- Multi-language Customer Support Intelligence.
- Personalized Healthcare Intervention Platform.
- Multi-modal Content Moderation Platform.
- Retail Traffic and Behavior Analytics.
- Energy Consumption Optimization Engine.
- Financial Fraud Detection System.
- Supply Chain Demand Forecasting Platform