SRK AI Loan Officer Advisor

The description states that SRK CAPITAL AI is an AI mortgage advisor. It functions as a conversational interface that:

• Provides Q&A grounded in real content with citations.
• Offers live, personalized rates derived from a pricing engine.
• Matches users to qualifying loan programs (e.g., conventional, FHA, VA).
• Guides users through a path to pre-approval.
• Uses semantic search over ~1,500 embedded chunks across ~780 documents.

It is built using Next.js, React, TypeScript, Hono edge API, Supabase/PostgreSQL with pgvector, OpenAI and Anthropic models, and integrates with a pricing engine via amortization formulas.

Inference The product appears to be an AI assistant for mortgage consumers that leverages retrieval-augmented generation (RAG) and structured data pipelines to deliver personalized loan insights in real time.

The description states the company’s tagline: “Skip the mortgage maze. An AI advisor that quotes live rates, matches you to the right loan, and gets you pre-approved—in minutes, not weeks.”

It also claims:

• The information already exists but is locked behind jargon and latency.
• A language model can be grounded in a corpus of mortgage content and connected to a live pricing engine.
• This turns a week-long process into one that answers in seconds.

Inference The positioning is centered on speed, transparency, and accessibility, targeting consumers who are frustrated by traditional mortgage processes. The evolution of the claim moves from identifying a problem (slow, opaque mortgage process) to proposing a solution (AI-driven conversational assistant with real-time data).

The description states:

• Users ask questions like “what’s the difference between a 7/1 ARM and a 30-year fixed for someone planning to move in 5 years?”
• The tool helps users understand loan options, qualify for programs, and walk them through pre-approval.
• It collects only what is needed, when it's needed.

Inference The target customer appears to be homebuyers or refinancers who are seeking clarity and speed in a complex process. The ICP likely includes individuals with basic financial literacy but limited knowledge of mortgage terminology, looking for personalized guidance without long delays.

There is no evidence provided about:

• Revenue streams
• Pricing models
• Monetization strategy
• Customer acquisition costs
• Any paid features or tiers

Not evidenced

The system uses:

• Semantic search with pgvector and text-embedding-3-small.
• A multi-model reasoning layer routing between GPT and Claude.
• A pricing engine computing effective rates using base sheet + LLPAs and amortization formulas.
• Guardrails including schema validation, contract harnesses (67 checks), CI failure detection, and model degrading gracefully.

Key technical details:

• Chunked documents indexed via vector search
• Retrieval runs as a Postgres RPC over pgvector
• Model routing based on task type
• Frontend/backend separation with automated contract verification

Inference The architecture shows an attempt at building a robust, scalable, and safe system using modern tools like RAG, vector databases, and multi-model orchestration. The team also addressed known issues such as incorrect ranking and silent failures in the ingestion pipeline.

There is no evidence of:

• Revenue
• Customers or users
• Product adoption
• Market traction
• Any form of monetization or business development beyond the hackathon submission

Not evidenced

The description does not mention competitors or market positioning relative to existing mortgage platforms, AI advisors, or fintech tools.

Not evidenced

1. No commercial traction: The product is described as a hackathon submission with no evidence of real-world usage.
2. Unproven business model: No indication of how the company intends to monetize or scale.
3. High technical complexity without validation: While the stack is sophisticated, there's no evidence that it works at scale or has been tested in production.
4. Single-person team: The project was built by one person (as stated), which raises questions about scalability and long-term maintenance.
5. Self-reported claims only: All descriptions are self-reported and unverified.

Inference This is a proof-of-concept or prototype, not a product with demonstrated market demand or commercial viability.

1. What specific user feedback did you gather during development?
2. How do you plan to monetize this tool once beyond the hackathon phase?
3. Have you validated the need for this solution in real-world scenarios?
4. Are there any partnerships or integrations with lenders or mortgage brokers already in place?
5. What are your plans for scaling beyond a single developer?
6. How do you intend to handle regulatory compliance and data privacy in the mortgage space?