75 hrs
reclaimed per cycle
E-commerce engagement
Phantasma · Tangible solutions for abstract problems.
We are consultancy, technology enablers for mid-market teams that have outgrown spreadsheets. We untangle heavy operations and help companies scale up by leveraging their existing data and infrastructure to build robust, Cloud native data pipelines, intelligence, CRM, and IoT systems. Quietly. With numbers behind every claim.
Manual processes and data silos — structured into reliable, connected pipelines that run without extra headcount.
Forecasting models, IoT networks, and cloud systems that turn raw data into autonomous, measurable decisions.
Short-term delivery or long-term partnership — expertise brought in without the cost of a full-time hire.
We audit operations, map where data is lost or misread, and document the actual problem before any code is written. Then we implement — or hand you the document so your team can.
Data warehouses, ML forecasting, IoT networks, API integrations, AI tools for non-technical users, BI dashboards, DevOps pipelines, e-commerce and billing automation.
Fixed-scope project with full handoff, or long-term: embedded consulting, stakeholder advisory, product management while the company scales.
Revenue grows but so does manual work — more orders, more spreadsheets, more staff doing the same tasks. We find the bottleneck, quantify it, and replace it with a system.
The data exists — in the ERP, CRM, IoT sensors, accounting system. But it lives in silos or no one can read it. We unify, clean, and surface the decisions it supports.
You need a data architect, ML engineer, or e-commerce specialist for a defined project. We step in, implement, document, and leave you with something you own.
How we work
Audit operations and model the ROI of automation.
Pipelines, data contracts, governance, and stack choices.
Pipelines shipped, validated, and tested against your history.
Documentation, training, runbooks. Your team owns it.
Optional retainer for evolution, monitoring, scaling.
Aggregate impact
75 hrs
reclaimed per cycle
E-commerce engagement
€15K
total cost over 2 years
Mining transformation
12 ha
autonomous, weather-aware irrigation
Precision viticulture
6 stores · 20 yrs
of data unified, GDP-compliant
Pharmacy chain
About Phantasma
Phantasma is a boutique technology enabler and process transformation consultancy. We work with mid-market teams — typically post-acquisition or post-spreadsheet — who need an architecture that scales without adding headcount and process nodes. We embed with your ops, build the system end-to-end, hand it back with documentation, and then leave.
We choose engagements where we can put a number against the outcome. If we cannot tell you what manual hours you will reclaim, what a cycle will cost to run, or what accuracy your forecasts will reach — we are not the right team for that project.
A 30-minute call is the quickest way to find out whether we can help.
30 minutes. We listen, you describe what's manual or operationally inefficient. You leave with a one-page outline of what an automation could look like, what tech brings to the process, what it would cost to run, and roughly what it would reclaim.
Or just email us: consult@solvephantasma.com
With Google, LinkedIn, or email and password.
A web platform, built in-house, that runs Facebook and Instagram advertising for product brands from one place: the product catalogue, a picture library, AI-generated ad pictures in the brand's own styles, a post composer, and campaign set-up with spending limits and approvals. The interface works in English and Georgian.
Small marketing teams spend most of their week on repetitive production: product photos, price banners, the same post in several formats, and a manual check before anything is paid for. General AI image tools help little in Georgia: they do not render Georgian script reliably and they ignore a brand's visual rules.
Products are imported from the shop's own Excel or CSV stock file (name, quantity, price in GEL); every row is checked and skipped rows are listed by their row number. Pictures live in a library with folders, tags and automatic clean-up of archived pictures after 30 days. The brand's picture styles are rules and options the team edits in the browser, imported from its prompt library in Excel. The composer assembles the prompt, sends the product photo and chosen style pictures to Google's Gemini image models, then draws the Georgian product name onto the picture itself, so the text is always exact. Campaigns are set up in a five-step wizard with budget limits, a typed confirmation and a log of every change, validated against Meta's Marketing API in dry-run mode before money is spent. Access is members-only and enforced in the database.
Products, pictures and campaigns sit in one searchable place. Georgian text on ad pictures is always correct. Nothing is published or spent without an explicit approval, and every change leaves a record. Built-in limits (30 AI pictures a day, 50 MB of library storage per person) keep running costs predictable.
Google Gemini image models, Meta Marketing API, Supabase (PostgreSQL, Auth, Storage, Edge Functions), row-level security, SheetJS, vanilla JavaScript.
A web platform for companies that bid in Georgian public procurement. It collects every tender in four categories (furniture, food, hygiene, construction materials) from the state procurement system, reads all of its documents, and turns them into requirements, deadlines, typical prices and competitor histories. It is built as a paid service: data is shown only after sign-in and is kept in private storage.
To find the few tenders worth bidding on, and the price that wins, a bidder has to read each tender's technical specification, price tables and contract draft, often several PDFs and spreadsheets per tender. Past winning prices are scattered across thousands of files. Doing this by hand for hundreds of tenders a month is not possible.
A resumable collector downloads tenders, bids, contracts and documents in small overnight batches. A document reader (PDF, Word, Excel) extracts each requirement, tags it (standards, certificates, laboratory tests), rates the tender's complexity and finds similar past tenders. Bidders' own price tables are read line by line and checked against the bid total. The site offers a dashboard, a tender list with filters, a page per tender, company and buyer pages with contract timelines, and a Products page with every bidder's line prices. Sign-in with Google; the data sits in private cloud storage.
900+ tenders, with the documents of 920 of them read (3,827 files, 28,642 requirement items, no failures); 442 companies and 517 buyers profiled. Checked against an expert's own labelling, the reader found 77 of 79 requirements in full and the other 2 in part. Product prices: 855 bidders' files from the last three months, 408 price tables read, 4,084 offer lines; 305 tables (75%) add up to the bid total.
Python collectors and document readers (PDF, Word, Excel), Supabase Auth and private Storage, Google sign-in, statically generated pages.
Personal Career Strategist is an online career service. The client signs in with Google, LinkedIn or email, uploads a CV and answers a preferences questionnaire. The platform reads the CV with AI, finds and scores vacancies, and prepares applications that are sent only after the client ticks them. A client dashboard and an admin panel with live chat run the service day to day.
Job seekers lose hours every week scanning job boards and rewriting the same CV, and many applications go to roles that never fitted. A service that does this for them has to be quick, show why a vacancy fits, and handle personal data with care.
CV reading: the text is taken out of the PDF in the browser and the name and contact details are removed first; only then does Claude (Anthropic) sort the rest into experience, education, skills and languages, and the client can turn it into a one-page CV (PDF or JPG). Matching: every vacancy gets a 0–100 score from six weighted factors (job title 45, skills 20, seniority 15, location 10, salary 5, language 5). The dashboard lists companies first with their matched vacancies, deadlines and status; applications go out only for vacancies the client approved. The admin panel shows every client's plan, searches and profile completeness, with live chat.
The service page shows the live funnel from vacancies found to applications sent. Personal data never reaches the AI model with the client's name or contacts attached, and no application goes out without the client's own approval.
Claude API (Anthropic), Supabase (Auth, PostgreSQL, Storage, Edge Functions), Google and LinkedIn sign-in, Python pipeline, Flitt hosted checkout, vanilla JavaScript.
A read-only monitoring pipeline for a German publisher of gifts and stationery that sells on amazon.de. Every day it pulls what Amazon currently shows for each product, stores a snapshot, logs every change, and writes the result into the team's Google Sheet, with an email summary or alert. It never changes anything on Amazon.
Listing content on Amazon can change without notice: titles, bullet points, images, A+ content and variations are altered by catalogue updates. The seller's team had no way to notice, and checking by hand in Seller Central was not repeatable for about 1,000 products.
Python on an existing Azure server, with a SQLite database that keeps daily snapshots and a permanent change log. Amazon's Selling Partner API supplies catalogue data, A+ content, seller reports and image dates; requests go in small batches to stay within Amazon's limits and resume automatically after a failure. A self-hosted n8n workflow runs the check every day at 18:00 Berlin time, writes a per-product tab and a per-SKU tab into the Google Sheet with the changes marked, and sends the email summary or alert.
935 products (1,873 SKUs) are checked in about 40 minutes per run, with all 935 completed and no failures in the last five recorded runs. A+ content is tracked on 780 of the 935 products. The team sees exactly what changed and when, at no running cost beyond the existing Azure server and Google Workspace.
Python, Amazon Selling Partner API, SQLite, self-hosted n8n, Google Sheets, Azure.
Comfort Option equips homes, apartment buildings and business centres with smart access: video doorbells, entrance doors, parking barriers and elevators, all opened from one mobile app. Our team designed and built the company's website (comfort-option.ge).
The product combines hardware, installation and an app, and it is bought by very different people: homeowners, residents of apartment buildings, business centres and developers. The site had to explain it quickly, in the languages its customers use, and lead every visitor to a concrete quote request.
One fast page in Georgian, English and Russian: an overview of the devices, a three-step how it works, a section per customer type, and live numbers from the client's own system (buildings, residents, devices), refreshed every hour. A short quote form asks for the project type, devices and number of flats or offices; phone, WhatsApp, Viber and Telegram links sit next to it. App Store and Google Play links, app screenshots, the privacy-policy and account-deletion pages the app stores require, and a sign-in for existing customers complete it.
The page stays current on its own: its counters read from the live system every hour, so no one updates them by hand. Every visitor, in any of the three languages, ends at the same short quote request or a messenger of their choice.
Trilingual website (Georgian, English, Russian), live counters from the client's system, quote request form, App Store and Google Play links.
MCG Market (mclub.ge) is a Georgian online marketplace for motorcycles, motorcycle parts and riding gear. It grew out of a Facebook community of about 70,000 members. Our team designed and built the platform.
Buying and selling motorcycles in Georgia happened in a large Facebook group: posts sank in the feed, nothing could be filtered by brand, year or engine size, and parts and gear were mixed in with bikes. The community needed one place that is easy to search and easy to post in.
A marketplace in three categories (motorcycles, parts, gear) with search and filters by make, model, location, condition, year, engine size and price, and a brand filter for 10 makes. Sellers sign in and post free listings with photos and details in five steps; buyers message sellers through the site. A shops directory, a blog with news and guides, a customs-duty calculator for imported motorcycles (by model year and engine size) and a contact form complete it. Built with Next.js on Vercel, with images in cloud storage. The platform takes no payments: buyers and sellers deal directly.
The community's trade moved from a social feed to a structured catalogue: every listing carries details the filters can search, and buyers reach sellers without leaving the platform. The customs calculator answers the import-duty question before a purchase.
Next.js, Vercel, Vercel Blob storage, on-site messaging, search and filters, Georgian interface.
Developed an end-to-end data automation ecosystem for a multi-channel e-commerce firm managing 8,000+ SKUs and 20+ manufacturers. Transformed a manual accounting process into a scalable €60/month cloud-native solution.
Variable pricing across 20 manufacturers, multi-channel marketplace commission complexity, logistical fragmentation, and a resource drain of 80 man-hours per cycle across 3 senior staff over 5 days.
Azure ETL/ELT pipeline with custom API retrieval platform, multi-format Data Lake handling JSON, CSV, and Parquet, star schema SQL Data Warehouse, Power App for logistics rate updates, Google Sheets-to-Azure pipeline for business users, and automated Paginated Reports generating client-ready invoices.
93.7% labor reduction. Turnaround time cut from 5 days to 2 days. Operational cost from 3 senior FTEs to €60/month. 75 hours of high-level employee time reclaimed per cycle.
Azure Data Factory, ADLS Gen2, MSSQL, Power BI, Power Apps, Amazon SP-API, Shopify API.
Integrated data ecosystem for a rapidly expanding mining and raw materials producer following a series of acquisitions. Automated document processing, unified task management with financial flows, and provided C-level executives with real-time decision-making snapshots.
Non-standardized processes across mining sites, data trapped in local computers and manual invoices, and massive human effort required to map cash flows to mining tasks.
Azure Data Factory and Databricks for ETL/ELT. Azure Machine Learning OCR to digitize manual invoices and map them to project tasks. ClickUp API integration linking every task to its financial cash flow. Power Apps for structured data entry. SharePoint as a governed document library. Row-Level Security and Active Directory integration.
Monthly task capacity 7× higher (40 to 280 tasks, +600%). Raw material processing +214% (3,500 to 11,000 units). 40% reduction in labor inefficiency. Data accessibility from under 10% to 80% organization-wide. Total cost of ownership €15,000 over two years.
Azure Databricks, Azure Data Factory, Azure Machine Learning, Power BI, Power Apps, SharePoint, ClickUp API.
Forensic audit and complete architectural overhaul of a failing e-commerce PHP monolith operating across Amazon and Shopify. Replaced with a modern decoupled microservices cloud architecture.
Single massive codebase caused regressions across modules. Only the original author could navigate the system. Reports hard-coded in Python. Single-threaded ETL lagged as sales volume grew.
Azure Data Factory for orchestration and ETL. Azure Databricks for Spark compute. Azure SQL as Single Source of Truth. Power BI for self-service BI. Azure DevOps CI/CD separating Dev, Staging, and Production. ADLS Gen2 Bronze layer for raw JSON. Terraform and Bicep Infrastructure as Code. Azure AD RBAC for department-level access control.
Azure Data Factory, Azure Databricks, Azure SQL, ADLS Gen2, Power BI, Azure DevOps, Terraform, Bicep, Azure Active Directory, ClickUp, Amazon SP-API, Shopify API.
Complete precision agriculture system for a 12-hectare vineyard combining IoT sensor infrastructure, edge computing, cloud data pipelines, and predictive automation for fully autonomous weather-aware irrigation.
IoT devices monitoring soil NPK, moisture, and temperature in real-time. Low-power always-on on-premise server as data retrieval center. Azure and Databricks pipelines syncing sensor data with Weather Forecast APIs. Machine learning model forecasting water needs based on evapotranspiration versus predicted rainfall. Automated pump control maintaining optimal moisture while minimizing energy.
IoT soil sensors, on-premise edge computing server, Azure cloud pipelines, Azure Databricks, Weather Forecast API, automated pump control systems.
End-to-end modernization of a 6-store pharmacy chain to align with Georgian healthcare regulations. Complex migration of 20 years of legacy data, AI-driven product mapping, and centralized IoT monitoring for GDP compliance.
Central Product Library integrating Ministry of Healthcare, Revenue Service, logistical databases, and historical invoices with AI-powered medicinal name matching. Custom Data Warehouse with automated procurement generating purchase orders per branch. Centralized IoT network monitoring temperature and humidity across 6 locations. Automated Ministry of Healthcare reporting workflows.
Custom Data Warehouse, IoT sensors, AI product mapping, ERP integration, Google Workspace, HubSpot.
End-to-end data pipeline and machine learning solution for a large-scale e-commerce distributor managing 12,000 SKUs across 200 brands. Achieved 98% forecasting accuracy, outperforming human procurement specialists at 81%.
Azure Data Factory and Databricks ETL/ELT pipelines. MSSQL Data Warehouse with Golden Record table unifying all sales channels. Feature engineering with rolling 7/30/90-day windows and YoY comparisons. Azure Machine Learning model generating automated replenishment suggestions. Rigorous backtesting against 4 years of historical data.
Azure Data Factory, Azure Databricks, MSSQL, Azure Machine Learning, Python, Shopify API, Amazon SP-API.
Turned a reactive century-old water utility serving 3M+ residents into a predictive data-driven operation. Daily incident throughput doubled, average fix time dropped 33%, high-severity pipe bursts fell 12% year-over-year.
PostgreSQL data warehouse consolidating SAP incident records, IoT telemetry from pressure gauges and reservoir levels, asset inventory, and billing consumption map. Machine learning forecasting achieving 80% daily, 90% weekly, and 93% monthly confidence. KPI and alert management layer connected back into SAP via custom REST API generating work orders. Power BI dashboards for incident volume, fix-time distribution, team utilization, and warehouse readiness.
SAP ERP, PostgreSQL, Python ML, Power BI, IoT telemetry, custom REST APIs, Azure infrastructure.