Applied AI & ML Engineering · Search, Discovery & Recommendations
RankTalos is a hands-on applied AI and ML engineering consultancy specializing in search relevance, hybrid semantic retrieval, recommendation systems, and evaluation harnesses. Test your live store in 60 seconds, commission a human-verified Gap Audit, or embed our engineers to ship the fix.
NL shopper queries failed on Furniture retailer A (Fast Simon); 9 returned zero results for in-stock items.
Verified October 31, 2026 against catalog truth
failed on Furniture retailer B (Algolia), plus 9 partial—returning dining chairs for “couch good for kids and dogs”.
72 pet-friendly sofas in stock
hybrid semantic & constraint retrieval on our live Talos Semantic Engine (usafurnish.com).
BM25 + dense vectors + constraint parser + LTR
of any Advisory Tier 1 Audit fee ($3,500 / $7,500–$5,000 / $10,000) credited toward a Tier 2 engineering build within 60 days.
4-Tier service & residency architecture
4-Tier Service & Pricing Architecture
Every engagement stands alone with clear deliverables or feeds the next tier—with 100% of any Advisory Tier 1 Audit fee credited toward a Tier 2 engineering build within 60 days.
Tier 0 · Self-Serve Diagnostic
Free (60s live check + 24h verified snapshot)
Enter your store URL to score Semantic & Constraint Recall (30%), Ranking Precision & Honesty (15%), Catalog Attribute Coverage (30%), and AI Crawler & Schema Readiness (25%), with a live side-by-side Talos Semantic Search preview.
Tier 1 · Productized Audits (100% Credited)
$3,500 / $7,500 Search · $5,000 / $10,000 Combined · $4,000 / $7,500 Recsys
Human-verified against your catalog truth table across 30–80 real shopper intents. Includes our $7,500 M&A Technical & Findability Due Diligence (5 business days, Independent Mode). $2,500 design-partner rate available for the first 3 Tier 1 Search Audits in exchange for named before/after case study rights.
Tier 2 · Fixed-Scope AI & ML Builds
$6,000–$12,000 Advisory · $12,000–$35,000 build (+ $1,500–$3,500/mo hosted) · $15,000–$40,000 Recsys
In-stack tuning (Algolia, Constructor, Fast Simon, Klevu, Shopify) or turnkey Talos hybrid search, two-tower & LTR recommendations ($15,000–$40,000), catalog enrichment ($6,000–$18,000), and NDCG@10 CI harnesses ($8,000–$18,000).
Tier 3 · Contract ML & Resident Teams
$4,000–$8,000/mo Principal · $16,000–$26,000/mo FT ($9,000–$14,000/mo HT) · $35,000–$55,000/mo Pod
Senior Search, Recommendation, and Applied ML Engineers embedded in your Slack, GitHub, and sprint ceremonies within 2–3 weeks, supervised weekly by a RankTalos Principal Architect.
Side-by-Side Proof · Lexical Search vs. Talos Semantic Engine
Standard keyword search and unconfigured third-party widgets fail whenever a shopper combines a category, a numeric limit, a use-case, or a negation. Compare actual results from Retailer A and Retailer B against our live Talos Semantic Engine reference implementation at usafurnish.com:
| Shopper Request | What Store Search Returned | Talos Constraint Parser | Talos Semantic Engine (usafurnish.com) |
|---|---|---|---|
| tv stand for 75 inch tv | 0 results (or unrelated popular-products fallback grid) | Category: Media Stand · Width ≥ 67" · In Stock | 10 matching solid-wood & oak media consoles (68"–84" W) ranked by TV-fit margin |
| couch that is good for kids and dogs | Dining chairs & counter stools matching the token "kids" | Category: Sofa / Couch · Fabric: Pet-Friendly / Stain-Resistant Performance | 72 performance-fabric & rub-tested sofas ranked by family durability |
| apartment size sofa under 80 inches that isn’t leather | 88" and 94" leather sectionals (ignores "< 80" and negation "isn’t") | Category: Sofa · Width ≤ 80" · Material ≠ Leather | 14 fabric & bouclé apartment sofas (68"–79" W) with zero leather leakage |
How the Findability Score (0–100) Works
Our scanner only flags a query as failed when your own structured catalog data contradicts the result. Every report decomposes your score across four weighted engineering pillars:
Weight: 30%
Numeric size/fit limits ("sofa under 80 inches"), price ceilings ("oak desk under $500"), material + attribute combinations, use-case ("good for kids and dogs"), and negation ("isn’t leather").
8 catalog-grounded queries where ≥ 2 in-stock items match; measures whether matching items appear in top 6 without zero-result failures.
Weight: 15%
Top-3 precision on multi-constraint queries and zero-result honesty check (does the engine admit 0 matches or silently dump unrelated best-sellers?).
Top-3 relevant matches divided by judgeable top-3 slots, with penalty for silent popular-product fallback grids when matches exist.
Weight: 30%
Structured vs. prose attribute extraction across top catalog categories (dimensions, materials, capacity, compatibility, SKU/GTIN).
Full credit (1.0) for attributes in structured options/tags/fields; half credit (0.5) for facts buried only in HTML prose; 0 for missing.
Weight: 25%
schema.org/Product JSON-LD completeness (price, availability, brand, variants, MPN/SKU, physical dimensions), XML sitemap, and robots.txt access for OAI-SearchBot, PerplexityBot, Googlebot, and Bingbot.
100-point rubric: JSON-LD (25), physical specs (15), offers (15), brand (10), variants (10), IDs (10), AI search crawlers (10), sitemap (5).
Who is RankTalos for? We serve two primary buyers: (1) E-commerce Directors and Founders at $5M–$50M spec-heavy stores (furniture, home, outdoor, lighting, cameras, audio, B2B catalogs) who start with the free Findability Score or a Tier 1 Audit, and (2) VPs of Engineering, CTOs, and Heads of ML at $20M–$250M retailers and marketplaces who hire us for Tier 2 search/recsys builds, evaluation harnesses, or Tier 3 Resident AI/ML Engineers.
How does the 100% Tier 1 Audit fee credit work? In Advisory mode (standard for merchant diagnostics), 100% of your Tier 1 Audit fee ($3,500 / $7,500 Search, $5,000 / $10,000 Combined, or $4,000 / $7,500 Recommendations) is credited toward any Tier 2 engineering build signed within 60 days of your readout call. We also offer a $2,500 design-partner rate for our first 3 Tier 1 Search Audits in exchange for named before/after case study rights.
What is the Talos Semantic Engine? Talos is our production hybrid retrieval and reranking engine (BM25 lexical + dense vector embeddings + numeric/unit constraint parser + cross-encoder reranker), live at usafurnish.com across 10,000+ furniture SKUs. We use it both as a live side-by-side benchmark inside our Findability Score tool and as an optional turnkey search layer in Tier 2.
How do Resident AI/ML Engineers differ from staffing marketplaces? Every RankTalos Resident Engineer ($16,000–$26,000/mo FT, or $9,000–$14,000/mo HT) comes pre-equipped with our evaluation harnesses and retrieval libraries, and includes weekly architecture and code review from a RankTalos Principal Architect. Engagements have a 3-month minimum, billed monthly upfront (Net-7), with a 20% first-year comp conversion buyout option after 6 months.
When do you use Independent Mode vs. Advisory Mode? Advisory Mode is standard for store audits where you may want us to build the fix. Independent Mode (where we bar ourselves from selling any Tier 2 or Tier 3 work to the tested store for 12 months) is reserved for our $7,500 M&A Technical & Findability Due-Diligence Package and formal vendor-selection RFPs.
Next Step · Engineering Architecture & Diagnostics
A 30-minute engineering call: your store, your search & recsys stack, and whether a Tier 1 Audit, Tier 2 Build, or Resident ML Engineer fits best. No slides, no obligation.