AI Tool Design for Domain Experts

The Problem I Solve

You can recognise quality in your field.

Good writing, coherent reasoning, proper procedure, valid claims — obvious when you see it, hard to articulate why.

You've wondered: could AI help scale this judgment?

The answer is yes. But not in the way most AI implementations attempt.

AI handles language.
Code handles judgment.

My Approach

Separate what AI does well from what requires human judgment.

I built Koher to encode 15 years of design teaching into an AI tool. The architecture shows the pattern:

Koher Architecture

DeBERTa

Analyses text, outputs confidence scores (0.0–1.0)

AI handles language pattern recognition — 98.38% accuracy

Rules

Converts confidence → severity levels

Code handles judgment thresholds — no AI opinion here

Claude API

Translates severity into plain language feedback

AI handles language generation — but only from verified inputs

The key insight: confidence scores never reach the language model directly. Every judgment passes through explicit, auditable rules.

Where This Applies

Any domain where experts recognise quality.

The architecture applies wherever:

• Experts can recognise quality but struggle to articulate why
• Training data is expensive — you can't just scrape the internet
• Wrong AI output is costly — hallucinations cause real harm
• Human judgment must remain central

Legal

Contract clause completeness, argument coherence

Education

Student concept clarity, assignment quality, learning outcome validation

Academic

Research proposal quality, literature review coverage

Financial

Investment thesis coherence, risk acknowledgment

Journalism

Source verification, claim-evidence alignment

Grant Writing

Proposal logic, budget justification, methodology clarity

Subscription Plans

Annual subscriptions. First month free.

Try for a month, commit for a year. Cancel during the trial with no obligation.

Advisory — ₹20,000/month

Annual subscription · Billed at ₹2,20,000/year after free trial

Ongoing guidance for teams building or planning domain-specific AI tools.

Up to 3 hours/month — including one scheduled 60-min call, async support, and quarterly strategy review

48-hour async response time
Architecture pattern selection
Implementation trade-off navigation

Best for: Teams with in-house developers who need expert direction

Architecture — ₹50,000/month

Annual subscription · Billed at ₹5,50,000/year after free trial

Hands-on architecture design and ongoing guidance:

Weekly consultation calls (60 min)
Architecture document creation and iteration
Training data strategy and annotation review
Rules layer design and threshold calibration
Implementation roadmap with milestones
Priority async support (24-hour response)

Best for: Organisations ready to build and needing architectural leadership

Development — ₹1,50,000/month

Annual subscription · Billed at ₹16,50,000/year after free trial

Full prototype development of a domain-specific AI validation tool:

Everything in Architecture plan, plus:
Data collection and annotation execution
Classifier training and evaluation
Rules layer implementation
API integration and UI design
Ongoing iteration and refinement
Dedicated support (same-day response)

Best for: Organisations that want a working AI tool delivered, not just designed

Background

What I bring.

Domain expertise 15+ years design education — NID, IIT Gandhinagar, Srishti, CEPT, Anant
ML/AI DeBERTa fine-tuning, multi-label classification, Anthropic API integration
Scale Installations at Mumbai Airport T2 (50M+ annual viewers)
Recognition TED India Fellow (2009), international residencies (Tokyo, Vienna, Rotterdam)

Featured project: Koher — DeBERTa classifier at 98.38% accuracy across 5 coherence dimensions.