Data & AI Platform Lead · Staff/Principal Data Engineer · Tbilisi
Archil
Chachanidze
I architect data and AI platforms: from undocumented operational systems to governed enterprise agents.
At Silk Development, I built production platforms spanning 49 SAP B1 companies and an undocumented 1,800-table Bitrix CRM on Airflow, MinIO, and ClickHouse. I now lead Unistream agent enablement and the final stage of a guarded Claude Enterprise rollout for roughly 40 users.
Source systems
1,800 Bitrix tables + 49 SAP B1 companies
Business model
Opaque structures rebuilt as usable entities
Data platform
Apache Airflow + MinIO + ClickHouse
Governed AI
Authoritative context + guarded rollout
Agentic Engineering
I architect the workflow; agents implement and verify
I design agentic delivery systems independently in Codex and Claude. I define the architecture, decomposition, context, constraints, and acceptance criteria; coder and QA roles implement, challenge, and correct the work. I retain final engineering judgment and approval.
- Architecture first: context, boundaries, and acceptance criteria are explicit before implementation.
- Independent QA: implementation and review remain separate roles with correction loops.
- Human acceptance: generated work is evidence to evaluate, not truth to trust blindly.
Experience
Apr 2026 – Present
Silk Development
Head of Data & Digital Apr 2026 – Present
Own the company's data-platform architecture, enterprise-AI implementation decisions, and technical roadmap.
- Architected an on-premises platform using Airflow, MinIO, and ClickHouse across 49 SAP Business One companies and a heavily customized Bitrix CRM.
- Reverse-engineered Bitrix's undocumented 1,800-table model and reconstructed its business entities and workflows into an analytics-ready warehouse.
- Built ingestion and custom change-detection pipelines around SAP B1's internal behavior, sustaining roughly 10-minute data freshness on modest hardware.
- Implemented the core data and agent-enablement layer connecting Unistream to company data; ongoing work expands authoritative-source coverage, business context, and supported workflows.
- Leading a guarded Claude Enterprise rollout for roughly 40 users, now in its final stage: training, adoption closeout, and safe access expansion.
Dec 2021 – Nov 2025
Making Science · Sweeft
Senior Data Engineer Nov 2024 – Nov 2025
Promoted after stabilizing a business-critical data platform and reducing failures from daily incidents to roughly one per quarter.
- Improved a key pipeline from 40% uptime to 99% within 8 months while keeping production live during the rebuild.
- Re-architected ETL orchestration for GA4, Adobe Analytics, and CM360 using BigQuery, Airflow, Cloud Functions, and Pub/Sub, reducing latency and restoring SLA performance.
- Established Python and Airflow implementation standards and onboarded new engineers into the platform.
Data Engineer Dec 2021 – Nov 2024 · 3 yrs
- Built and supported large-scale ETL/ELT pipelines for Adobe Analytics, GA4, CM360, and Power BI exports on GCP.
- Migrated 100+ legacy scripts into Airflow and Cloud Scheduler with retries, monitoring, and SLA controls.
- Introduced Pydantic validation and schema normalization to reduce drift and improve downstream data quality.
- Automated BigQuery loads for Looker Studio so analysts had fresher reporting data without manual requests.
Selected Projects
Chaos-injectable synthetic data engine on FastAPI. Generates deterministic, YAML-schema-defined data, then injects controlled failures so the systems consuming it can be tested against realistic chaos.
- 19 chaos operations: schema drift, latency, encoding corruption, late arrivals, partial loads, header anomalies, and more.
- Cross-schema correlation: the same
customer_idstays consistent across orders, payments, and events via Redis-backed pools. - Low-latency delivery: Redis-backed pre-generation feeds WebSocket streams, while LLM-assisted tools author valid schemas and chaos profiles.
7,278items/sec peak (single host)
19chaos operations
30-cellreproducible bench matrix
Marketing Analytics Platform Redesign
Redesigned analytics infrastructure for three teams with different source systems, consolidating ingestion into contract-driven Airflow pipelines on GCP. Result: 99% uptime, 1M+ rows processed per day, and one production failure across a year of operation.
99%uptime
1M+rows/day
50production DAGs
Technical Scope
Data & AI Platforms
ClickHouse
on-premises analytics, fast ingestion, and warehouse serving
MinIO
S3-compatible lake storage and reproducible data layers
Enterprise AI
platform rollout, sensitive-data guardrails, training, and adoption
SAP Business One
multi-company ingestion, normalization, and change detection
Bitrix CRM
undocumented-model reverse engineering and business reconstruction
LLM & MCP Systems
coder-QA orchestration, agent harnesses, tool integration, local-model experiments, and evaluation
Engineering Core
Python
production ETL pipelines and data services
SQL
data modeling and query performance tuning
BigQuery
partitioned tables, analytics workloads, large-scale querying
Apache Airflow
orchestration, scheduling, retries, and SLA management
ETL / ELT
idempotent and schema-aware pipeline design
FastAPI
data APIs and contract-first service development
GCP Services
Pub/Sub
event-driven ingestion and decoupled workflows
Cloud Functions
serverless ETL and lightweight automation
Cloud Run
containerized services and scheduled workloads
GCS
staging, archival, and data lake storage
Data Engineering Practices
Data Contracts
Pydantic validation and schema drift control
Pipeline Architecture
modular, configuration-driven systems
Idempotent Processing
safe reruns, replayability, and operational reliability
Storage, APIs, and Delivery
PostgreSQL
relational modeling and transactional systems
Redis
caching, stateful workflows, and fast lookups
Docker
reproducible services and containerized pipelines
Pandas
profiling, transformation, and exploratory analysis
Infrastructure
Terraform
GCP provisioning and environment setup
CI/CD
GitHub Actions and automated deployment workflows
Kubernetes
production operations and service deployment
Prometheus + Grafana
monitoring, alerting, and service visibility
Working Knowledge
Apache Spark
familiar with local development and core concepts
dbt
comfortable with transformation patterns
Kafka
tested locally and understand event-streaming basics
Snowflake
querying and analytical usage
Databricks
notebooks and platform fundamentals
AWS / Azure
working familiarity outside primary GCP experience