ls cases/

Ten projects that show what we do

Client names are under NDA, so we talk about the industry, the scale and the measured result: faithfulness, latency, cost, conversion, uptime. A reference call with a team we have worked with is available on request.

AI cases

AI integration and LLM systems

Where the client asks for an "AI capability" and gets a measurable speed-up, lower cost, and quality that can be proven on a golden set.

Fintech / RAG · compliance2.4M documents

RAG copilot for compliance analysts

Task. Analysts spent 30–60 minutes on each case review, searching 2.4M internal documents and checking regulatory requirements. Open-source search returned irrelevant results, and naive RAG hallucinated.

What we did. Hybrid search (BM25 + embeddings) with cross-encoder re-ranking. Chunking that follows document structure. A separate fact-checking pass on Claude. A citation-aware UI where every claim links to its source. Evaluation on a golden set of 1200 pairs.

Result. Faithfulness 0.94 (against a 0.71 baseline), case review time down from 40 to 6 minutes, token cost 72% lower thanks to prefix caching and routing simple questions to cheaper models.

  • Claude
  • pgvector
  • BM25
  • Ragas
  • Langfuse
6.5×faster analysis
0.94faithfulness
−72%cost vs naive RAG
E-commerce / agents800K tickets / month

AI agent in a marketplace support team

Task. A high volume of standard requests: "where is my order", "refund me", "change the address". First-response time was 9 minutes, support staff were burning out, NPS stood at −18.

What we did. An agent pipeline on LangGraph with tool use: order status check, return initiation, address update, escalation. Human-in-the-loop for refunds above a threshold. A separate safety layer against prompt injection.

Result. First response went from 9 minutes to 12 seconds, 68% of requests are resolved automatically, support NPS is +12. The support team moved to complex cases and proactive outreach.

  • GPT-4o
  • LangGraph
  • tool-use
  • Postgres
  • Redis
45×faster first response
68%auto-resolution
+12 NPSsupport
SaaS / LLMOps10M req/day · multi-team

Production LLM gateway with routing and caching

Task. 14 product teams were calling OpenAI and Anthropic independently. Cost grew faster than revenue, nobody could see who was spending what, rate-limit failures were frequent and there was no fallback.

What we did. Designed and launched a gateway: a unified API, routing between Claude, GPT and Llama by task type and SLO, a semantic cache, per-tenant rate limiting, retry and fallback on 429 / 5xx, cost attribution via OpenTelemetry. PII filters and a per-user token budget.

Result. Token cost down 64% in a quarter. Gateway overhead of 42 ms at p99. Zero rate-limit incidents in 6 months. Every team sees its spend in real time.

  • Go
  • Redis
  • vLLM
  • OpenTelemetry
  • Langfuse
10Mrequests / day
−64%token cost
42 msoverhead p99
DevOps / AI agent1.4K incidents / month

AI agent for automatic triage of production incidents

Task. The SRE team was drowning in alerts: 50% false positives, MTTA of 18 minutes, on-call engineers woken 3–4 times a night.

What we did. An AI agent reads the alert, pulls the related metrics and logs over MCP, searches past incidents in postmortems (RAG), and proposes a hypothesis and a runbook. Below a confidence threshold it escalates to a human with the context already gathered.

Result. MTTA went from 18 minutes to 2 minutes, because the agent gathers the context for the engineer. 41% of alerts are closed without on-call as known issues. Night-time wake-ups are down 3.5×.

  • Claude
  • MCP
  • Prometheus
  • Loki
  • pgvector
9×lower MTTA
41%auto-triage
−72%on-call burden
Analytics / data lake & DWH120K call hours / month

LLM-powered analysis of sales calls

Task. Call recordings were unstructured noise. Listening to them manually was impossible, the legacy BI covered only the CRM, and managers filled in fields inconsistently.

What we did. A pipeline: transcription (Whisper), diarization, structured extraction (an LLM writing to a strict JSON schema), highlights of objections and key questions, ClickHouse views, dashboards for the sales team and for product.

Result. 100% of calls are tagged automatically within 5 minutes of the call ending. The product team sees real customer objections, and managers get feedback with quotes. Sales conversion grew 9% in a quarter.

  • Whisper
  • Claude
  • ClickHouse
  • Kafka
100%call coverage
5 minSLA from end of call
+9%conversion
High-load cases

Architecture and engineering for systems under load

Payments, RTB, search, multi-tenant SaaS, real-time sessions. The metrics are what these projects were built for.

Fintech / payments35K RPS · 4 regions

Payment platform: from a monolith to an event-driven core

Task. The payment platform hit a database bottleneck at peak, releases went out once every two weeks and came with incidents, and the business was planning 5–6× growth in transactions.

What we did. Decomposed the monolith into 9 services around bounded contexts. Kafka as the backbone, outbox and saga orchestration. A multi-region Kubernetes cluster with an active-active payment core.

Result. Payment API p99 went from 850 ms to 110 ms. Releases are daily. Infrastructure cost is down 40%. The platform sustained 6× growth.

  • Go
  • Kafka
  • Postgres
  • K8s
  • AWS
7×lower p99
6×TPS growth
−40%cloud cost
AdTech / RTB250K QPS · multi-AZ

Real-time bidding pipeline with a 95 ms SLA

Task. Launch an RTB bidder that fits into the 100 ms auction budget and reliably handles 200K+ QPS at peak.

What we did. A pipeline in Go with an in-memory feature store on Redis Cluster. Analytics and model training in ClickHouse, real-time ingestion through Kafka. Multi-AZ deployment with health-aware load balancing.

Result. 250K QPS, with auctions completing in 95 ms at p99. 18 months without a single incident that cost revenue.

  • Go
  • Redis
  • ClickHouse
  • Kafka
  • AWS
250KQPS
95 msSLA p99
99.99%uptime
E-commerce / search12M SKU · 4M MAU

Search and faceted navigation for a marketplace

Task. Search across 12M products was slow, at 800 ms p95, and collapsed under peak load. Search-to-cart conversion had stalled.

What we did. Migrated to Elasticsearch with pre-calculated facets and autocomplete, split the nodes into data, coordinating and ingest roles, and moved indexing to a stream from Kafka.

Result. Search p95 went from 800 ms to 80 ms. Search-to-cart conversion is up 18%. A full catalog reindex takes 25 minutes instead of 6 hours.

  • Elasticsearch
  • Node.js
  • Kafka
  • Postgres
10×faster p95
+18%conversion
14×faster indexing
B2B SaaS / multi-tenant1.2K tenants

Multi-tenant platform: isolation and scalability

Task. The SaaS had outgrown its shared-database approach: noisy neighbours affected each other, compliance audits took weeks, and engineers provisioned tenants by hand.

What we did. Moved to schema-per-tenant with automatic provisioning through Terraform. Separated the critical path from analytics. Per-tenant rate limits and resource quotas in Kubernetes.

Result. Query p95 is 3× lower. Tenant provisioning went from 2 days to 90 seconds. Audit and compliance are automated.

  • Postgres
  • Terraform
  • K8s
  • Go
3×lower p95
90 sprovisioning
0incidents in 12 months
Gaming / real-time800K MAU · <50 ms p99

Real-time backend for multiplayer sessions

Task. A backend for real-time sessions with predictable latency that copes with seasonal audience spikes.

What we did. Matchmaking and state services in Go with Redis Streams. Sessions are distributed across regional shards, so the failure of one shard does not touch the others. Blue/green deployments with warm-up.

Result. Network latency p99 is under 50 ms. 800K MAU without degradation. 18 months in production with zero downtime.

  • Go
  • Redis Streams
  • NATS
  • K8s
<50 msp99
800KMAU
0downtime in 18 months
Next step

Have a similar project?

Tell us about your system and its metrics. We will show how we would approach it and what it would cost.

Discuss a project