Case · Helm (self-applied) · 12 weeks initial · ongoing Helm retainer · Customer-Zero for Content Army
Internal — Talpro Universe · 24/7 AI content engine (proof-of-concept for Content Army)
Helm (self-applied) · 12 weeks initial · ongoing Helm retainer · Customer-Zero for Content Army
Talpro Universe (content engine)
Outcomes · measured
Surfaces published into
competitorx, Talpro India, HCI Talks, etc · 29 pieces in 14 weeks
Cost compression
vs equivalent human content team · 4× compression
Brand voice eval score
vs 0.62 baseline pre-gating
Narrative
01 · Problem
Problem
Talpro Universe runs ~12 brand surfaces that need content cadence — competitorx.cloud Insights, Talpro India blog, HCI Talks newsletter, HireIQ Pro changelogs, Jharokha customer education, etc. A traditional content team for 12 surfaces would cost ~₹85L/year (writer + editor + SEO + designer per 3-surface cluster) and still produce thin work because no human writer has end-to-end context across HR-staffing, fintech-ops, smart-doorbell, and AI-eval-harness simultaneously. The hypothesis: an AI agent system with the institutional context already in our memory MCP could draft, fact-check, eval, and ship content across all 12 surfaces — reviewed by one human editor at one-tenth the cost. The challenge: prove it without producing content slop that erodes the brand.
02 · Approach
Approach
Twelve-week Helm-shaped engagement (internal, ongoing). Week 1: define the eval harness for content quality — fact-density, source-traceability, voice-match-to-brand, AI-disclosure-discipline (every JAYA-authored piece carries a transparency badge with model + draft-time + editor name). Weeks 2–4: build JAYA agents per surface — a research agent that pulls from Talpro memory MCP + web search, a draft agent (Claude Opus for headlines, Sonnet for body), a fact-check agent (independent eval that re-derives every numerical claim), an SEO agent. Week 5–8: ship to four surfaces in production with weekly editor review and a transparency dashboard at /transparency. Weeks 9–12: scale to all 12 surfaces, tune the editor-loop to ~3 hours / week / surface (vs ~15 hours / week / surface for a human writer).
03 · Outcome
Outcome
JAYA has published 29 pieces across 12 brand surfaces in 14 weeks of operation, with 100% transparency badge coverage (every piece carries model, draft-time, editor). Editor time: ~36 hours / week across all 12 surfaces (one human, ~3 hours per surface per week). Equivalent human-team cost would have been ₹85L/year for similar coverage; JAYA actual cost runs at ₹22L/year all-in (inference + editor + tools) — a 4× cost compression. Brand voice consistency on the eval-harness scores 0.86 across all surfaces (vs 0.62 baseline before eval gating). JAYA is the Customer-Zero proof for the Content Army offering at /agent-army/content-army — every prospect call references this exact case study.
04 · In their words
“JAYA is the running proof that an AI Content Army works. Twelve surfaces. One editor. The cost of one senior writer’s annual salary.”
05 — Who led this engagement
Bhaskar Anand. Every first call.
Founder & CEO, CompetitorX. Pune, India. No associate-level handoff — the person who led this engagement is the person who takes your scoping call.