FactMiner

Real-time speech-to-fact capture, entity extraction, and claim verification pipeline.

code View on GitHub

⚡ The Prototype // Idea #3

FactMiner

Built in 2.0h with uv, Python, FastAPI, WebSockets, OpenAI API, Wikipedia REST API

code View on GitHub

tips_and_updates The 5 Morning Sparks

Bringing graphics to comminication

IDEA #1 lightbulb

Word cloud of a chat log

Classic way of reducing dimensionality of text

It's a classic way of reducing dimensionality. Just would be great if I can get stats about the chat log, strip the metadata, and visualise a word cloud for specific participants or everyone at once.

Data Viz, Social

IDEA #2 lightbulb

Conversation based graphics

Image/video generation based on the chat

AI generation of the content from the conversation.

Generative Art, AI

IDEA #3 lightbulb

Fact capture and check

data mine the audio

Speech to text, into content concept extraction, into fact checking/explaining. e.g. I want to capture someone mentioning Marseilles population as a fact. The system captures this as a fact mentioned, checks the real number and labels the fact. Same for difficult words, named entities. Fact checking is not always required, but if someone mentions e.g. Sartre, it captures and overview with information pertinent to the context.

Voice, AI, Data SHIPPED

IDEA #4 lightbulb

Sentiment tips

Sentiment alignment based on audio

Analyze the speech content and tone on emotional/sentiment spectrum. Then suggest which part of the speech need work based on stated goal. e.g. if I want to sound angry, help me develop content and delivery based on how it is now, so that it makes my delivery evaluate to more angry.

Voice, AI

IDEA #5 lightbulb

Knowledge Graph of conversation

Improving those horrible Facilitator

The current conversation summarisers are not very good and they don't persist the important knowledge outside of the output. I am interested in capturing information from a chat and organising it into a knowledge graph, checking it against currently existing one, so I can actually make meetings productive

Data Viz, AI

construction

How It Was Built With AI in 2.0 Hours

One-shot standalone build

Spun up the spec as a standalone repo with a modern UI, wired up Wikipedia and OpenAI fact-checking APIs, and added live mic mining plus sample test scenarios

thumb_up

What Rocked

✓Wikipedia ground truth and simulated scenarios worked straight out of the box without needing an API key.

bug_report

What Broke & Fixed

✗Python 3.14 alpha segfaulted on pydantic-core, so had to pin to Python 3.13.

smart_toy AI Interaction Summary

Genesis Prompt: "/spec-implementation"
Interaction Turns: 5 human turns across 126 execution steps

Key Guidance Turns:
  • Turn 2: This should be standalone new repository, preferably in python 3.14, modern ui setup. We should connect it to some API which al...
  • Turn 3: souonds good
  • Turn 4: just change the tone to be in line with how my previous entries sound. Keep it shorter, the process steps should really just be...
  • Turn 5: yes

Tools Executed: run_command (24), write_to_file (17), view_file (10), grep_search (3), replace_file_content (3), find_by_name (1)
Files Modified: .gitignore, .python-version, AGENTS.md, README.md, SKILL.md, __init__.py, db.py, fact_checker.py