Learn AI from five real products.
Not toy demos. Each level is a feature running today on a real website, built for a real business need. Every case study shows how it works, what broke when we tested it, and the code-level fix.
The one rule behind all five
The model talks.
Our code decides.
A language model is good at understanding a question and writing a clear answer. It is bad at being exact. So the facts (scores, prices, statuses, people) always come from our own data and code. The model only explains them.
Each level adds one technique
Read them in order: every level reuses ideas from the one below it.
- Level 1 SellYourLogic · AI Interview Coach Make a model return data, not prose An AI coach writes interview questions from a resume and grades every answer. The model always replies in a fixed JSON shape that code can check, and every score is computed by code. Structured output and grading Read the case study →
- Level 2 Logictote Kids Lab · Ask Baby Tote Look things up before you answer Children ask science questions. Before the model answers, our code finds the few lesson notes that are about the question and gives the model only those, so answers stay true, short and inside what our labs teach. Retrieval-augmented generation (RAG) Read the case study →
- Level 3 PakDigit Markets · Rehnuma, the markets assistant Check every number before it is shown Visitors ask about today's dollar, gold and petrol rates, Shariah status and zakat. A pipeline answers from live data first, then from our guides, and only then asks a model, whose every number is checked against our data. RAG + live data + number checking + voice Read the case study →
- Level 4 FindPropertyDealers · Rehnuma, the property guide Let the model choose the tools, and verify what it did A visitor types "5 marla plot near Islamabad airport, up to 25 lakh". The model chooses which of our functions to call (search societies, find dealers, draft a requirement), reads the results and decides the next step. Our code builds every card. AI agent with tool calling, memory and human confirmation Read the case study →
- Level 5 SellYourLogic · Expert Finder (Logi) One AI engine, two very different jobs A company describes a need, "a data analyst for 2 weeks, under $100/h", and gets a shortlist of real experts. Same agent pattern as level 4, running on the engine SellYourLogic already had for the Interview Coach. An agent on a shared engine, with search by meaning over people Read the case study →
At a glance
How the five compare.
| Level 1 SellYourLogic | Level 2 Logictote Kids Lab | Level 3 PakDigit Markets | Level 4 FindPropertyDealers | Level 5 SellYourLogic | |
|---|---|---|---|---|---|
| Main technique | Structured output | RAG | RAG + data + checking | Agent + tools | Agent on a shared engine |
| Where facts come from | Code computes scores | Our lesson notes | Our database + guides | Tool results | Tool results |
| Search by meaning | Question bank | Lesson notes | Our guides | Not yet | Expert profiles |
| Remembers the chat | No | No | No | Yes | Yes |
| Takes an action | No | No | No | Drafts a requirement | Drafts a hire request |
| Voice | Spoken answers | Read aloud | Urdu + English input | Not yet | Not yet |
| Without AI | Bank + quick checks | Closest note | Data answers | Guided buttons | Keyword match |
| Measured with | Human-graded set | 48-question eval | Tests + feedback | 12-case eval | 12-case eval + traces |
Shared building blocks
The same four habits appear in every product.
A list of providers
An ordered list of AI services with on/off, test and fallback, plus one switch that pauses all AI.
Local first
Our own models answer first. Hosted services are only a fallback, and visitors are told when one is in use.
Privacy by default
What people type is not stored. We keep daily totals only: asked, answered, refused.
Measure before changing
Tests and evaluation sets run before a model, prompt or threshold changes.
Words you will meet
- Language model
- A program trained on text that predicts what to write next. Good with words, unreliable with exact facts.
- Prompt
- The instructions and context we send to a model with each request.
- Structured output
- Asking the model to reply in a fixed shape (JSON) that code can check.
- Embedding
- A list of numbers (a vector) that captures the meaning of a text. Similar meanings, similar vectors.
- Cosine similarity
- How closely two vectors point the same way, from −1 to 1. Our "how alike is the meaning" score.
- RAG
- Retrieval-augmented generation: look up the relevant notes first, then let the model answer from them.
- Grounding
- Tying an answer to our own data, and checking it against that data.
- Tool calling
- The model asks to run one of our functions with arguments; our code runs it and returns the result.
- Agent
- A model in a loop that chooses tools, reads results and decides the next step towards a goal.
- Evaluation set
- Fixed test questions with expected results, used to measure a model or a change before shipping it.
- Provider
- A service that runs a model: our own computer, or a hosted company such as Google, OpenAI or Anthropic.
- Prompt injection
- Text that tries to give the model new instructions. We treat all outside text as data.
AI Playground
Change chunk size, similarity threshold, model and prompt on our case studies, and see what was retrieved, the exact prompt, the answer, the speed and the cost.
Open the Playground →Live workshops
Hands-on sessions that build each level step by step, hosted by our founder and by experts on SellYourLogic.