How the Anselm Project Builds AI-Powered Biblical Character Studies
Anselm Project builds AI-powered biblical character studies with a five-expert system, phased workflow, and synthesis to create rigorous, distinct portraits.
I built a new engine for biblical character studies. Here's how it works.
The Problem with AI Character Analysis
Ask an AI to write about Moses and it'll give you generic Sunday School material. Ask it five times and you'll get the same opening sentence with different words. AI loves patterns, and biblical characters have obvious patterns.
I needed to break that cycle.
The Five Expert System
Instead of one AI call trying to be an expert in everything, I split character analysis into five specialized perspectives. Each expert has one job and stays in their lane.
The Biographer
This expert catalogs the character's life chronologically. But it works in two phases.
Phase A catalogs every event. Major events, minor events, timeline markers. Just the facts from Scripture.
Phase B takes that event catalog and writes the life narrative. It analyzes major events for theological significance. The two-phase approach prevents the AI from inventing events while analyzing them.
The Relational Analyst
This expert focuses exclusively on relationships. God-relationship first, then family dynamics, allies, enemies. The prompt explicitly tells the AI to focus on quality and theological significance, not just listing names.
The Canonical Theologian
This expert maps where the character appears across Scripture and how later biblical authors interpret them. It distinguishes between narrative appearance and theological reference.
David appears in 1-2 Samuel narratively. But the Psalms reference him, Chronicles reinterprets him, and the New Testament invokes him messianically. Different functions.
The Typologist
This expert identifies how Old Testament figures prefigure Christ. The prompt forbids allegory and forced connections. If there's no clear typology, say so briefly and move on.
The Tradition Scholar
This expert pulls from Josephus, rabbinic literature, Philo, the Targums, apocrypha, pseudepigrapha, Dead Sea Scrolls, and Church Fathers. It names specific sources and distinguishes historical detail from embellishment.
The Three-Phase Workflow
Phase 1: Sequential Event Cataloging
The Biographer Phase A runs first. Alone. It catalogs events chronologically and identifies major versus minor events.
This phase must complete before anything else happens because Biographer Phase B needs the event catalog.
Phase 2: Massive Parallel Fan-Out
Once Phase A completes, everything else launches simultaneously. Five parallel API calls:
- Biographer Phase B (life narrative + event analysis)
- Relational Analyst (all relationships)
- Canonical Theologian (Scripture's witness)
- Typologist (Christological patterns)
- Tradition Scholar (extra-biblical material)
All five experts see the same character name but receive completely different prompts. The morphology expert from biblical reports doesn't exist here because we're analyzing people, not parsing Greek verbs.
This parallel execution cuts processing time significantly. Instead of waiting for five sequential calls, they all run simultaneously.
Phase 3: Master Synthesis
After all expert analyses complete, a sixth AI call synthesizes everything into the master introduction.
The synthesizer receives context from all five experts and writes four sections:
- Identity statement (2-3 sentences: who they are)
- Historical context (one paragraph placing them in history)
- Theological significance (one paragraph on why they matter)
- Character summary (400-500 word unified portrait)
The character summary isn't just concatenated expert analyses. It's a synthesized essay that weaves biographical, relational, canonical, typological, and traditional dimensions into flowing prose.
The Anti-Redundancy System
I built explicit style rules to prevent repetitive AI language.
Each expert gets seven opening sentence patterns to choose from. Different patterns for different experts. The Biographer might open with "Scripture introduces {name} in {book}..." while the Relational Analyst opens with "{name} stood before Yahweh as..."
Each expert also gets forbidden patterns. The AI is explicitly told never to write "The canonical scope of {name}..." or "{name}'s relationship with God can be characterized as..." Generic AI phrases get blocked.
Tone guidance tells each expert how to approach their analysis. Sentence variety rules force alternation between simple, compound, and complex sentences.
Character-Specific Scaling
Major characters like Moses or Paul get comprehensive treatment. 3-5 paragraphs per section. Rich detail.
Minor characters like Melchizedek get concise treatment. 1-2 paragraphs per section. If you have little to say, say little.
The system determines prominence and adjusts scope instructions accordingly.
Validation First
Before any of this happens, a gatekeeper validates the character name.
It checks three things:
- Is this actually a biblical character?
- Is the name ambiguous? (Saul could be King Saul or Saul of Tarsus)
- What's the canonical name to use?
If the name is ambiguous, the system asks for clarification before proceeding. No point generating a report on "Mary" when there are five different Marys in Scripture.
Token Economics
A full character study makes seven API calls to start:
- 1 validation call (gatekeeper)
- 1 sequential call (Biographer Phase A)
- 5 parallel calls (Phase 2 experts)
- 1 synthesis call (master introduction)
The Output Structure
The final report includes:
- Identity statement, historical context, theological significance, character summary
- Full biographical timeline and event catalog
- Life narrative with major event analyses
- Complete relational analysis (God, family, allies, enemies)
- Canonical presence mapping
- Typological assessment
- Extra-biblical tradition summary
- Processing metadata (tokens, costs, API calls)
Everything is structured JSON with specific content blocks for consistent formatting.
Why This Works
The mixture-of-experts architecture prevents hallucinations. Each expert has a narrow focus. The Biographer doesn't try to do typology. The Typologist doesn't try to do textual criticism.
The two-phase biographical approach prevents event invention. Catalog first, analyze second.
The anti-redundancy system forces stylistic variation. Different openings, forbidden patterns, tone guidance.
The parallel execution makes it fast. Five experts run simultaneously instead of sequentially.
The validation prevents garbage input.
It's the fourth engine I built for the Anselm Project, and like the biblical reports engine, it works because the architecture matches the task.
If you want to see what these character studies actually look like, check the Share Gallery or generate one. They only cost 1 credit - same as the normal reports.
God bless, everyone.
Related Articles
How the Anselm Project Generates Biblical Studies
Explore how the Anselm Project generates biblical studies reports using a mixture-of-experts AI pipe...
How Anselm's Topical Reports Produce Accurate Biblical Theology
Anselm's topical reports deliver accurate biblical theology: a mixture-of-experts that scans the can...
Genesis 3:15 Explained: Protoevangelium, Hebrew Grammar, and Messianic Seed
Explore Genesis 3:15 as the protoevangelium with grammatical, theological, and canonical readings th...