This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend
What I Built
I built NQT Saathi, a local, offline prep buddy for TCS NQT, for my batchmate [FRIEND_NAME] at UCET, Vinoba Bhave University, Hazaribagh.
There are no on-campus placement drives at our college, so TCS NQT is the main way in for both of us. His problems are very real:
- His internet is patchy and mobile data runs out fast, so ChatGPT-style tools are unreliable for daily practice.
- He understands concepts better in Hinglish than in textbook English, and almost every prep resource is English-only.
- His laptop is low-end with no GPU.
- He doesn't want to pay for another subscription.
So NQT Saathi runs entirely on his laptop, with no internet and no cost:
- Hinglish explanations: paste any aptitude, reasoning, or verbal question and get a step-by-step explanation in simple Hinglish.
- Coding hints, not spoilers: a three-level hint ladder (nudge, bigger hint, approach). Full code only if he asks.
- Mock test mode: timed sections that mimic the NQT pattern, with a score at the end.
- Weak-topic tracker: every wrong answer is logged in a local SQLite file, and the next practice set leans toward his weak topics.
- Daily revision queue: a small spaced-revision list of the mistakes he made earlier.
His scores, mistakes, and study habits never leave his machine.
Demo
Code
Pagdandi
Take the path less scrolled.
Pagdandi (पगडंडी, “footpath”) is a mobile-first, audio-led walk companion. Pick 15, 30, or 60 minutes and a mood; open-weight Gemma creates a handful of weather-aware observation missions, reads them aloud, then helps turn the walk into a private field note. The key outcome is intentionally not engagement: Pagdandi shows time outside vs. time the screen was active.
Built October 5–11, 2026 for the DEV Hacktoberfest Open-Source AI Challenge: Week 1 — Touch Grass.
Screenshots
Add final home, walk-mode, and field-journal screenshots before submission.
| Plan a walk | Pocket-first walk mode | A private field note |
|---|---|---|
docs/screenshots/home.png |
docs/screenshots/walk.png |
docs/screenshots/journal.png |
Why open matters
Gemma is not decoration here: it creates the quests, cautious photo guesses, and final field notes. A small provider interface makes those same open weights available in two ways:
-
gemini-apiserves Gemma through Google’s Gemini API for the public Render demo. -
ollamaruns…
How I Built It
| Layer | Choice |
|---|---|
| Model runtime | Ollama, fully local inference |
| Explanations and aptitude | Qwen2.5 (3B for low RAM, 7B if the machine can handle it), open weights, Apache 2.0 |
| Coding hints | Qwen2.5-Coder, open weights |
| Backend | Python + FastAPI |
| Frontend | Simple web UI served locally |
| Storage | SQLite (progress, mistakes, revision queue) |
| Question bank | Plain JSON files, easy to edit and extend |
The project is built around the open model, not just calling it:
- A system prompt tuned for Hinglish teaching. The model explains like a senior who cleared NQT: simple Hinglish, small steps.
- A hint ladder for coding. The prompt enforces levels so the model doesn't dump the answer on the first try.
- Swappable models. One config line switches between the 3B and 7B model, depending on his laptop.
- Everything is a plain file. Questions are JSON and progress is SQLite, so he can back up the whole thing by copying a folder.
Why Does Open Innovation Matter?
- It works with no internet. Once the model is pulled, NQT Saathi runs in a hostel room with no network. A hosted API fails exactly when he needs practice most.
- It costs nothing to run. No API key, no quota, no billing. He can ask 500 doubts a day and it is still free.
- His data stays with him. His weak spots and mock scores never touch a server he doesn't control.
- I could change how the model behaves. With open weights and an open runtime, I could swap models per task, tune the prompts for Hinglish, and size the model to his hardware instead of accepting one vendor's choices.
- It outlives me. If a closed API changes pricing or retires a model, the tool breaks. Open weights on his laptop keep working.
What he said
I installed it on his laptop and sat with him for a session.
[ADD HIS REAL REACTION HERE, in his own words]
What he used most: [ADD, e.g. Hinglish explanations / mock test mode].
What he asked me to change: [ADD, e.g. more verbal ability questions].
Top comments (0)