Merciven
Work · Applied AI

Truth estimation and research search

An API that scores how likely a statement is to be true, using a research corpus, and shows its evidence.

Organisation
Spiritual Data / Neodocto (US nonprofit, remote)
My role
Senior Analytics Developer
When
2023 to 2025
Stack
Python, NLP, Machine learning, Semantic search

Problem statement

The organisation's research lived in thousands of accounts and papers. Researchers needed to know what the evidence says about a claim, and find related work, without reading everything.

Objectives & goals

  • Score a statement against the corpus
  • Show which studies moved the score
  • Classify incoming accounts and papers
  • Recommend related research

Process

  1. DiscoverWorked with researchers on the questions they asked most.
  2. DefineEvidence weighting by study design, peer review and sample size.
  3. BuildIngestion and NLP classification pipeline, truth-estimation API, semantic search and recommendations, executive dashboards.
  4. ProveChecked scores against claims researchers had already assessed.

Who it's for

Researcher

Checks a claim and reads the evidence.

Leadership

Follows the research through dashboards.

Challenges

  • A single case report shouldn't outweigh a large study
  • Negated findings ("did not improve") flip the meaning
  • When there's no evidence, the answer must say so

What I built

Evidence-weighted score

Log-odds from a 50% prior, weighted by design, peer review and sample size.

Transparent output

Every study's stance, relevance and contribution.

Semantic search

Find studies by meaning, not keywords.

Related research

Recommendations by similarity.

Flow

  1. Statement
  2. Retrieve studies
  3. Detect stance
  4. Weight by quality
  5. Combine log-odds
  6. Score + evidence

Output

{"statement": "Daily coffee intake improves long-term memory performance",
 "probability": 0.04, "verdict": "Likely false",
 "evidence": [{"id": "S6", "stance": "contradicts", "weight": 1.0, ...}]}
The sample API on a synthetic corpus of 8 invented studies.

Results

Scoredwith evidence shown
NLPclassification pipeline

Tools & languages

  • Python
  • NLP
  • Machine learning
  • Semantic search
  • Data pipelines
  • Dashboards

The code

Retrieval, stance, evidence weighting and a JSON API on the standard library. 5 tests.

The repository is private. I'm happy to share it with a hiring team or walk you through it on a call.

Request a code walkthrough

Samples are written with synthetic data. No client code or client data is used.