1C · LLM · semantic retrieval
Matching a customer request to a 1C product catalog
A practical case combining an LLM, semantic retrieval and manager verification when customer wording differs from the catalog.
Read on vc.ru (RU)Architecture_Doc //
Engineering approach
to automation
From the ancient mechanics of Heron of Alexandria to the digital singularity: we build business automation that turns process chaos into a strict, predictable algorithm.
// Mission_Statement
We build precise software constructs that lift the operational load off people.
> EXECUTE: scale_operations();
Scroll
Engineering > AI Hype
Reliable automation starts with a defined input, a verifiable result, known exceptions and an acceptable error rate. We add a model only where deterministic rules are not enough.
Phase_01 // Decomposition
Decomposition
We break chaotic business processes into atomic, measurable machine operations.
Phase_02 // Structuring
Structuring
Data engineering and strict OCR parsing of unpredictable corporate documents.
Phase_03 // Hybrid Core
Hybrid Core
A model extracts meaning, while code verifies identifiers, amounts, statuses and permissions. Uncertain results are routed to a person.
RESULT //We combine a model, business rules, source data and an action log into one testable system.
Data & Permissions // 0x01
Architecture for
your constraints
Deployment depends on the data and the required actions. We separate public and private sources, grant the agent minimum permissions and keep high-impact actions behind explicit confirmation.
Deployment options
Cloud services, private infrastructure and local models are evaluated against quality, cost and data requirements.
Source separation
Public price lists and restricted documents receive separate access policies, indexes and audit trails.
Controlled web access
External search is enabled only for workflows that need it. Results retain their source and remain subject to verification.
[ ACCESS IS EXPLICIT ]
The agent receives only the data and tools required for one defined workflow.
Authority & Values
01
Transparent ROI
Before development, we record volume, staff time, error cost and exception rate. The pilot is measured against that baseline.
02
Vendor Independence
We select cloud or local models based on quality, budget and data requirements. The architecture should not depend on one vendor without a reason.
03
Full-Cycle Integration
From server setup and data engineering to web control panels and training your staff to operate the system.
Field Notes // Cases & Research
Cases, experiments and reviews
Client cases are listed separately from experiments and reviews of third-party technology. The original articles are currently in Russian.
All Geron Labs casesGoogle Sheets · Telegram WebApp
Using Google Sheets as the first database for a fitness product
Fit Bot keeps plans and client data in spreadsheets while a Telegram WebApp provides the working interface.
Read on vc.ru (RU)SaaS · Telegram · Cloudflare R2
Field staff dispatch and photo reports
A cleaning operations case covering dispatch, task statuses, Telegram and protected photo storage, including technical trade-offs.
Read on vc.ru (RU)Avito · MoySklad · Telegram
Connecting Avito, MoySklad and a Telegram operations hub
An AI assistant verifies product data and conversation context while order events and human follow-up remain separate actions.
Read on vc.ru (RU)SMS · GSM · local gateway
When an AI model is not needed
A local Android SMS gateway solved a narrow GSM relay task more directly than an AI component would have.
Read on vc.ru (RU)Technology review · not a client case
MiroFish: a review of a multi-agent simulation project
A review of a third-party research project in which multiple agents simulate how a community reacts. This is not a Geron Labs client case.
Read on vc.ru (RU)MVP · Farcaster · NFT
A Farcaster experiment with personalized NFTs
A small MVP that turned Farcaster profile data into a personalized NFT, listed separately from business automation cases.
Read on vc.ru (RU)Geron Labs // Team
The people behind the work
Founder · CEO
Sergey Sidorov
Runs Geron Labs: strategy, key clients and the quality of delivery. He is personally involved in every project — mapping the client process, designing the solution architecture and signing off the work before launch.
since 2025 — Geron Labs: AI agents and business process automation for B2B.
2023–2024 — Python and aiogram, parsing, browser automation with Selenium, C# and ZennoPoster.
2021–2024 — blockchain infrastructure: BSC, Polygon and Solana nodes, Solidity smart contracts.
Sales · AI Video
Daniil Ponomarev
Daniil leads sales: first contact with clients, scoping the request and guiding it to a signed contract. He also runs the visual practice — AI-generated ads and product videos, with examples on the AI video production page. Razgon is his own project.
Projects · Web3 Advisor
Konstantin Shevel
Konstantin manages projects: timelines, milestones, client communication and acceptance. As a Web3 and blockchain advisor he guides the team on the market and the strategy of the blockchain practice and opens access to partners in the industry.
SYS: COOKIE
We use technical cookies (Cloudflare) so the site works correctly. Privacy Policy