Georgia’s renewable-energy mix is beginning to diversify beyond hydropower, while the electricity market itself is gradually becoming more sophisticated.

For Georgian energy-tech startup Vynity, both developments create a growing need for better forecasting.

The company uses data analytics and artificial intelligence to forecast generation from hydropower, solar and wind plants. It also develops tools for imbalance forecasting, market intelligence and commercial decision-making.

Vynity says it currently has 35 renewable-energy plants under model and has been working with paying customers since April 2025.

In an interview with renewables.GE, co-founder Giorgi Jalagonia explains why the company moved beyond hydropower forecasting, where forecasting errors create real financial costs, why wind is currently particularly difficult to predict in Georgia, and why he believes forecasting will eventually become an underlying layer of the electricity market.

From forecasting generation to forecasting commercial exposure

renewables.GE: Vynity started by focusing heavily on hydropower forecasting. Since then, the platform has expanded toward solar, wind, consumption forecasting and market analytics. What changed?

Giorgi Jalagonia: Two things changed: one in the market, and one in how we understood the problem.

The market reason is simple. When we started, Georgia was effectively a hydro system. Hydropower still produces roughly 11 TWh of about 13.8 TWh of national generation, so if you wanted to forecast anything that mattered here, you forecast water.

But the pipeline of wind and solar projects that had been announced for years began turning into actual connection agreements and construction.

Anyone building a forecasting business only for hydro was building for the system as it was rather than the system it is becoming.

The deeper reason is that we were wrong about what we were selling.

We thought we were selling a hydro forecast. We were actually selling a position: the producer’s net exposure at settlement.

A producer does not care about the hydro number in isolation. They care about the gap between what they told the system operator they would deliver and what they actually delivered, priced at the balancing price.

Once you frame it that way, hydro generation is only one term in the equation. There may also be the producer’s own consumption, a solar plant elsewhere in the portfolio, and the price at which the difference will be settled.

The expansion therefore was not diversification for its own sake. It was closing the loop on the commercial question the customer is trying to answer: not “how much will I generate?” but “what should I declare, and what will it cost me if I’m wrong?”

renewables.GE: Renewable forecasting is often discussed as a technical accuracy problem. From the producer’s perspective, where does inaccurate forecasting actually hurt financially or operationally?

Giorgi Jalagonia: It hurts in four places, and only the first is obvious.

The first is imbalance settlement. In Georgia, the structure is asymmetric. If a producer over-forecasts and needs to cover a shortfall, they may be buying electricity at a much higher price. If they under-forecast and sell surplus electricity into balancing, they may receive a much lower price.

You can lose on both sides of the error.

The second is foregone contracting. This one is less visible because it never appears directly on the P&L.

A producer who does not trust their own forward view contracts conservatively. They sell less bilaterally than they potentially could, or they avoid participating in the exchange because the downside of making a commitment they cannot meet feels greater than the potential upside.

The third area is operations.

For a reservoir hydro plant, the forecast affects dispatch and drawdown timing. Getting the inflow view wrong by a day or two can mean spilling water that should have been stored, or storing water that could have been sold during a higher-price period.

For run-of-river plants, forecasting can also influence maintenance scheduling.

The fourth area is staff time.

Small and mid-sized producers in Georgia often have one person preparing the forecast manually each day, sometimes the chief engineer or commercial director, alongside regulatory submissions.

That is not free, and it is not always reproducible or auditable. When that person is on holiday, forecast quality can drop.

Why wind is currently the hardest to forecast

renewables.GE: Vynity provides forecasts across hydro, solar and wind. Which of these technologies is currently the hardest to forecast accurately in Georgia, and why?

Giorgi Jalagonia: The answer depends on the forecast horizon.

Intraday to day-ahead, per unit of installed capacity, I would say wind.

Power is roughly proportional to the cube of wind speed, so relatively small errors in wind-speed forecasts can translate into much larger errors in output.

Georgia’s terrain makes this harder. Many attractive wind sites are attractive precisely because of channelled, terrain-driven airflow, which coarse global weather models can struggle to resolve.

There is also very little operating history in Georgia. The installed wind fleet is still small, so you cannot rely on the same depth of local operational data that exists in markets such as Germany or Turkey.

For multi-day to seasonal forecasting, hydro becomes more difficult.

Many catchments are ungauged or sparsely gauged. Snowpack, which determines a large part of the spring and early-summer profile, is also poorly observed on the ground, so you rely more heavily on satellite and reanalysis products.

There is another Georgia-specific complication: part of “hydrology” is actually upstream human decision-making. If a cascade upstream changes its release schedule, no meteorological model will tell you that.

Solar is generally the easiest of the three under clear conditions. The difficult periods are convective cloud on summer afternoons and rapid ramps around frontal passages.

If I had to name one: wind, today.

Ask me again in three years, when there is a larger operating wind fleet producing the historical data that makes the problem more tractable.

renewables.GE: What data does Vynity need from a power plant before it can start producing useful forecasts? How much depends on plant history compared with external data such as weather and hydrology?

Giorgi Jalagonia: We deliberately keep the minimum viable input small.

We need the plant location and coordinates, installed capacity, technology type and key technical parameters. For wind, that might be the turbine curve. For solar, inverter and panel configuration. For hydro, head and design flow.

Historical generation at hourly resolution is also important, together with operating constraints that do not appear in public datasets, such as scheduled outages, minimum-flow obligations or curtailment agreements.

The balance between plant history and external data changes with the horizon.

For the next few hours, the plant’s recent behaviour and current state carry a lot of the signal.

For day-ahead and the following few days, external information becomes more important — particularly numerical weather prediction and, for hydro, upstream hydrological conditions.

Further out, external data dominates, while plant history serves mainly as calibration.

But that understates what plant history actually does.

Weather models tell you what the atmosphere will do. Plant history tells you what this specific machine does with it.

Every plant has an idiosyncratic transfer function. A turbine may underperform its theoretical curve at partial load. An intake may behave differently at certain flows. A site’s measured wind may consistently run above or below a gridded weather model.

That mapping has to be learned from data, and a lot of the accuracy comes from there.

Software first, deeper integration later

renewables.GE: Vynity is software-based and does not require on-site hardware integration. What advantages does that provide, and what are the limitations?

Giorgi Jalagonia: The advantages are commercial as much as technical.

Deployment can be measured in days rather than quarters. There is no procurement cycle for sensors, no installation window and no need to coordinate with the plant’s electrical contractor before beginning.

For a mid-sized producer with a small technical team, that can make the difference between “we will consider it next budget year” and “let’s try it.”

It also means we can serve plants that are too small to justify additional hardware investment. In Georgia, that is a significant part of the generation fleet and a segment that has traditionally been underserved.

The limitations are real, and I would rather state them clearly.

A system with a live SCADA feed sees things we may see later or have to infer: an unplanned trip, a derated unit, an inverter offline or an intake blocked with debris.

Over the next few hours, that operational state can be a large part of the answer. A fully integrated system therefore has a real advantage at very short horizons.

There are also activities that require sub-minute data, such as fast ancillary services, where our current approach is not designed to operate.

Our view is that this is a sequencing choice rather than a permanent constraint.

Zero-integration gets you in the door and lets you demonstrate value quickly. If a customer later wants deeper integration, we can ingest a SCADA or historian feed.

The point is that integration is optional and earned, not a precondition.

renewables.GE: How do you measure whether a forecast is actually good enough to create commercial value?

Giorgi Jalagonia: Accuracy metrics such as MAE and nMAE matter, but they are proxies.

Two forecasts with identical statistical accuracy can have very different commercial value because errors are not priced equally.

An error during a high-balancing-price hour may cost many times more than the same error during a low-price hour. Over- and under-delivery may also be priced differently.

So we look at three levels.

The first is statistical: how the forecast performs against installed capacity and, importantly, against whatever the customer was doing before.

The relevant question is not simply “our error is X%.” It is “your error was X% and is now Y%.”

The second is financial. We re-price the forecast error against the actual balancing prices for that period. That converts an improvement in accuracy into GEL.

The third is behavioural.

Does the forecast change what the customer is willing to do? Do they contract more volume forward? Do they participate more actively in the market? Do they stop keeping a manual safety buffer?

That can create value beyond the settlement line.

I would also be careful about claiming universal accuracy improvements.

A well-run plant with a competent forecaster and a stable reservoir is a difficult baseline to beat significantly.

Where we consistently see more opportunity is in run-of-river assets, portfolios, plants without dedicated forecasting staff and the ramp or tail events that manual approaches tend to handle poorly.

Georgia’s market still weakens the incentive to forecast hourly

renewables.GE: How does Georgia’s current electricity-market structure affect demand for better forecasting tools?

Giorgi Jalagonia: It creates demand because balancing is enormous.

A significant share of electricity delivery passes through the balancing mechanism, and the difference between balancing prices can create substantial exposure for producers.

But the current market structure also suppresses demand because balancing is still settled monthly rather than hourly.

Georgia has been operating under a transitional market model. Day-ahead and intraday markets exist through GENEX, but participation remains limited, while the obligation to balance on an hourly basis has been postponed.

Under monthly netting, individual hourly errors are partly forgiven because over- and under-deliveries can offset one another within the month before settlement.

That materially weakens the incentive to forecast each hour accurately.

When people ask why forecasting is not already ubiquitous in Georgia, that is a major part of the answer.

The market has not yet fully priced the thing.

What changes with hourly settlement is not incremental; it is a step change.

Once every hour is priced on its own, the offsetting disappears. A producer’s exposure increases, and forecasting moves from being a useful tool toward becoming a cost of doing business.

We are building for the market that is coming rather than the one that exists.

That is a deliberate bet, and it has a cost. It makes today’s sales conversation harder because you are partly selling against a future price signal.

But when the rules change, we want to already have the plants, operating history and customer trust.

Building that history takes years, and you cannot compress it once the deadline arrives.

renewables.GE: Vynity also automates regulatory forecasting documents. Is this mainly a convenience, or can it materially change how smaller producers participate in the market?

Giorgi Jalagonia: For a large producer with a commercial department, automating forecast submissions is mainly an efficiency gain: fewer manual hours, fewer errors and a better audit trail.

For a small producer, it can be more significant.

The administrative burden of market participation is largely fixed. It does not scale down with plant size.

Preparing declarations, meeting submission deadlines and keeping formats correct can require almost the same administrative effort for a 5 MW plant as for a 50 MW plant, but the smaller plant has far less revenue over which to spread that cost.

That quietly determines who participates.

Small producers may remain on default arrangements not because those arrangements are necessarily optimal, but because the overhead of doing something else is too high.

Removing that overhead does not automatically turn a small producer into an active trader. The forecast still has to be good enough, and the market still has to provide sufficient value.

But it removes one barrier that matters more than it may initially appear.

From forecasting tool to market infrastructure

renewables.GE: As solar and wind capacity grows in Georgia, do you expect forecasting to remain mainly a tool for individual producers?

Giorgi Jalagonia: It becomes infrastructure.

That is not really a prediction. It is what has happened in markets that liberalised before Georgia.

Forecasting remains a standalone product only while imbalance is the main thing being optimised.

Once you have a liquid short-term market, forecasting becomes an input into something else.

It becomes an input to a trading decision, a battery’s charge and discharge schedule, an aggregator’s portfolio position or a balancing responsible party’s net position.

The forecast stops being the product and becomes the layer everyone else builds on.

Portfolio effects then become increasingly important.

A single producer forecasting alone is exposed to its own full error.

Put 20 producers into an aggregated portfolio and part of those errors can cancel each other because they are imperfectly correlated.

That creates a real economic gain for whoever operates the portfolio, which is why aggregation and forecasting tend to converge.

Storage makes this even more important.

A battery’s commercial value depends heavily on forecast quality because you are monetising the difference between the price now and the price later.

A better view of “later” becomes extremely valuable.

So yes, I see producers, traders, aggregators, storage operators and eventually balancing-service providers consuming the same forecasting layer for different purposes.

Whether Vynity becomes that layer in Georgia depends on execution, but the shape of the end state seems fairly determined.

Where AI could change the energy sector next

renewables.GE: Beyond forecasting, where do you think AI could have the biggest real impact on the energy sector over the next five years?

Giorgi Jalagonia: Four areas, in rough order of how confident I am.

The first is weather modelling itself.

AI weather models such as GraphCast, Aurora and ECMWF’s AIFS represent one of the biggest changes in the underlying inputs to this industry in decades.

They can run much faster and at far lower computational cost than traditional approaches. That makes large ensembles economically feasible for many more users.

That does not only improve forecasts. It changes who can produce them.

The second area is grid infrastructure and connection queues.

Across Europe, the binding constraint on renewable deployment is increasingly the interconnection queue and the network-study backlog rather than the availability of turbines or panels.

That is partly a document and simulation problem — exactly the type of work where current AI systems can help.

Compressing multi-year study timelines could potentially unlock substantial clean-energy capacity.

Third is demand-side flexibility.

Millions of EVs, heat pumps and home batteries are becoming controllable resources. Coordinating them requires continuous, high-dimensional optimisation under uncertainty.

That is a natural fit for AI.

The fourth is asset inspection and predictive maintenance.

This is less glamorous, but it is already working: computer vision using drone imagery of wind-turbine blades, solar panels and power lines, or acoustic and vibration analysis on rotating equipment.

One area I would describe as overhyped is fully autonomous AI trading of physical energy.

The models may be capable enough, but the regulatory and risk-governance frameworks are nowhere close.

For longer than the current enthusiasm suggests, I expect decision support with a human in the loop.

Taking Vynity beyond Georgia

renewables.GE: Which parts of Vynity can move easily into another electricity market, and which need to be rebuilt locally?

Giorgi Jalagonia: The split is roughly 60/40 in our favour.

The core modelling transfers relatively easily.

The underlying physics and machine-learning approaches are portable. A turbine power curve behaves the same way in Georgia and Romania. The methodology for bias-correcting global weather models is also transferable.

The plant-level learning approach transfers, as does the basic product experience: portfolio views, alerts and the workflow of a producer’s day.

Other things have to be rebuilt.

Settlement and imbalance rules are country-specific. Every market prices forecasting errors differently.

Regulatory submission formats and deadlines are local.

Market-data integration is also local because every TSO and exchange publishes data differently.

Hydrological modelling is particularly difficult to transfer. Catchment behaviour, snowmelt regimes, upstream cascade operations and measurement networks vary significantly between regions.

Wind and solar models transfer more easily, which is one reason those markets are a natural first step outward.

Language and commercial relationships matter too. Energy sales are relationship sales, and relationships are local.

Our expansion logic is therefore to move first into adjacent markets where the structural conditions are similar and the financial incentive for accurate forecasting is strong.

Priority number one is Turkey.

We have already found some early traction there and expect to share more concrete details on partnerships and expansion soon.

Building an energy-tech company in Georgia

renewables.GE: Vynity received a GITA grant and later participated in dena’s SET Tech Festival. What did those programmes change for the company?

Giorgi Jalagonia: In 2024 we won a GEL 150,000 GITA grant.

It was the first real milestone in what we were trying to build, and it was worth more than the money itself.

The funding covered modelling expenses and helped us access data and licences we needed.

But it also marked us out as a promising Georgian startup, which sent a clear signal to potential clients and counterparties.

dena’s SET Tech Festival gave us something different: confidence that we could hold our own against much bigger players.

Meeting major energy companies, learning from them and seeing how they think has shaped our approach to entering new markets.

renewables.GE: From your experience building an energy-tech company in Georgia, what has been harder than expected?

Giorgi Jalagonia: Not the technology.

Founders are usually expected to say the technology was the mountain. It was not.

The modelling is difficult and we work hard at it, but it is a known kind of difficult. There is academic literature, there are benchmarks and you can measure whether you are improving.

The hardest part has been the sales cycle into a conservative sector.

Energy assets are long-lived and capital-intensive. The people operating them have asymmetric professional risk.

Nobody gets promoted for adopting a forecasting platform, but somebody gets blamed if a new system contributes to an incident.

So the rational move for the individual you are pitching is often to wait.

Overcoming that is not simply a matter of producing a better demo.

You may need to run alongside the customer’s existing process for months and let the results accumulate until they are comfortable.

That patience is expensive for a small company.

Data access is a close second.

We need historical generation data, and producers often treat it as commercially sensitive.

Even once they agree to share it, the data may be incomplete, stored in old spreadsheets, have gaps or unclear timestamps.

A meaningful part of the early customer work becomes what I call “data archaeology.”

Capital is difficult in a different way.

Georgia has grants and some early-stage support, but it does not yet have a deep-tech venture ecosystem.

For international investors, being a Georgian company can also raise immediate questions around market size, jurisdiction and potential exit opportunities.

The result is that grants are easier to access than equity.

But grants are not always the right instrument once a company has customers and revenue and needs time and hiring capacity to scale.

What was easier than expected was technical hiring and the willingness of energy-sector professionals to engage once we were actually in the room with them.

Georgian energy professionals are not technophobic.

They are risk-averse, which is different — and reasonable.

Once a plant manager sees a week of our forecasts sitting next to their own and ours are better, the conversation changes completely.

Getting to that week is the hard part.

Source note: Interview with Giorgi Jalagonia, Co-Founder of Vynity. Company data, market figures and operational examples cited in the interview are based on his responses and materials provided by Vynity.