5 production case studies

AI agent case studies, with the numbers.

LangGraph multi-agent systems, LLM cost optimization and production AI automations, shipped for paying clients. Each one shows the problem, the architecture and what changed after launch.

Case studies
5 in production
Industries
E-commerce, SaaS, Sales
Build time
8–18 weeks each
Stack
LangGraph, LangChain
  1. 01 / 0595.2%Routing precision on 211 labelled prompts (from 80.3%)5,140Rows in the largest approved import6Independent safeguards on every write2,987Automated tests guarding the agent
    • ERP / Operations
    • Dubai-based client · SaaS for Odoo users
    • 18 weeks

    ERP Operations Agent - Letting an LLM Write to Odoo Safely

    A chat-first operations layer for Odoo ERP that turns a sentence or a messy spreadsheet into approved, verified writes, with six safeguards that each assume the model is wrong.

    • LangGraph
    • Odoo ERP
    • Human-in-the-Loop
    • Multi-Tenant SaaS
    Read the case study Service: AI agent development
  2. 02 / 0520×Cheaper per AI pitch after capping thinking tokens~20 ptsGross margin protected on each paid tier50×Lower monthly cost of a free user, by design5Security issues found and fixed before launch
    • Creator economy / SaaS
    • US-based creator-economy startup
    • 13 weeks

    AI Pitch Engine for UGC Creators - 20x Cheaper AI, Same Quality

    A three-app SaaS that helps UGC creators find verified brand contacts, write cold pitches that don't read like AI, send them from their own mailbox on a schedule, and pay through a credit and subscription stack whose prices come from measured costs.

    • Gemini
    • FastAPI
    • LLM Cost Optimization
    • Stripe Billing
    Read the case study Service: LLM cost optimization
  3. 03 / 0565%Reduction in monthly LLM spend31%Semantic cache hit rate+0.2CSAT delta (within noise)220msImprovement in P50 response time
    • SaaS
    • High-traffic B2B SaaS
    • 8 weeks

    LLM Cost Optimization for SaaS Platform

    Cut OpenAI API spend by 65% on a high-volume SaaS without a single-point drop in CSAT - through semantic caching, prompt compression, and intent-aware model routing.

    • Cost Optimization
    • Semantic Caching
    • Prompt Compression
    • Model Routing
    Read the case study Service: LLM cost optimization
  4. 04 / 05~2 minAverage response time on qualified leads (from 26 hrs)3.1xIncrease in SQL-to-opportunity conversion+27%Lift in pipeline-sourced revenue per SDR45%Reduction in SDR research time
    • Sales / RevOps
    • Series B B2B software company
    • 9 weeks

    AI Lead Qualification for B2B Sales

    Built a LangChain-powered lead-scoring agent that enriches inbound leads, scores fit and intent, and routes high-priority opportunities straight to AEs - cutting response time from 26 hours to under 2 minutes.

    • LangChain
    • RAG
    • Salesforce
    • Slack
    Read the case study Service: LangChain engineering
  5. 05 / 05~75%Fewer input tokens per draft from hybrid retrieval90%Cheaper static context with model-side caching3Independent checks before any message auto-sends1,077Automated tests across 102 test modules
    • B2B Sales / Outbound
    • Workly (my own product)
    • Ongoing

    Workly - A Self-Improving Multi-Agent Outbound Engine

    Fifteen specialised AI agents research each B2B prospect, write to them on the channel they actually answer, review their own work before it sends, and improve their own prompts from what comes back.

    • Multi-Agent
    • Self-Improving Agents
    • RAG
    • LinkedIn + Email
    Read the case study Service: AI agent development
Every build above ran on the same 6-phase playbookDiscovery, architecture, build, evals, gated rollout, optimization. See the real code.

Want results like these? Start with one call.

Book a free scoping call
What happens after you book
  1. A 30-minute call about your workflow
  2. A one-page scope with price and timeline, free
  3. You decide: build, or keep the scope and walk away