Cellium. Biotech
SPRIND Challenge 2026 · Engineering Microbial Consortia
Applicant profile · Straubing, Germany

Jet fuel from Bavarian waste, made by consortia that stay in control.

We turn variable dairy and food residues into medium-chain acids, the building blocks for sustainable aviation fuel (SAF), with a digital twin driven by artificial intelligence (AI) holding the microbial community in balance.

The Cellium platform
Physical twin Proprietary lab-scale bioreactor

Sensors, actuators and data acquisition designed around what the twin needs.

Digital twin Live, learning model of the run

Predicts hours ahead, tests adjustments in simulation, then acts.

Standard lab bioreactor
3–5 Cellium, by design
50+
Closed-loop AI control Virtual experiments & test planning Scale-up prediction Learns from every run
Cellium so far

From a university thesis to an AI-native bioreactor in two years.

  1. Origin

    Master's thesis

    At the Technical University of Munich (TUM): 3D-printed bioreactors for bioplastic fermentation. EU patent application filed.

  2. 2025

    XPLORE & Chemstars

    Accelerators at UnternehmerTUM. ~80 customer interviews.

  3. 2025

    First paid project

    Custom bioreactor for a client who then secured €2M in funding.

  4. Dec 2025

    Own platform starts

    Funded by that project. Reactor and twin built as one system.

  5. 2026

    MVP ready

    Minimum viable product: live lab fermentation feeding real-time data into the twin and its machine-learning models.

  6. 2026

    TUM IDEAward finals

    Finalist in the 2026 TUM IDEAward for research-based startup ideas.

  7. Nov 2026

    SPRIND Challenge

    Consortia for jet-fuel precursors from local waste.

  8. 2027 · planned

    Market-ready product

    Industrial manufacturing partners, EU product certification, company founding.

1stAI-native bioreactor: hardware and software built in parallel to unlock the power of automation
10×more data per experiment than a standard lab bioreactor
100%owned by the team: €60K pre-seed from paid projects and our own equity
5industry and academic partners
The demand

Aviation must decarbonise, and Europe can't source enough sustainable fuel.

187.6 Mtmillion tonnes of CO₂ emitted by European flights in 2024
2.8%share of sustainable aviation fuel in the 39.3 million tonnes of jet fuel supplied in the EU in 2025
3×price of aviation biofuel vs fossil jet: €1,925/t vs €640/t
Minimum SAF share required by EU law (ReFuelEU Aviation)
Where the EU's 2025 SAF feedstock came from

85% imported, and 80% of it is used cooking oil. Plants operating today reach about 1.5 Mt/yr by 2030, short of the ~2.1 Mt the 6% mandate needs.

The feedstock

The carbon is already here, in Lower Bavaria's residue streams.

4.76 Mtwhey and whey products from Bavarian dairies in 2025
10.9 Mtfood waste in Germany in 2023, 17% of it from food processing
3.77 MtCO₂-equivalent avoided by the 1.1 million tonnes of SAF used in the EU in 2025
  1. SourceDairies & food processorsWhey, food residues
  2. CelliumMicrobial consortiumSugars → lactate → caproate
  3. UpgradingJet-range hydrocarbonsMedium-chain acids to SAF
  4. UseJet fuel blendMixed into conventional kerosene
The product

Medium-chain fatty acids: the bridge from waste to jet fuel.

Chain elongation, two carbons per cycle
C2acetateC4butyrateC6caproateaviation lubricant estersC8caprylatebio-based lubricants & oilsC10caprateour target:aviation biofuel+2 C+2 C+2 C+2 C
What they are
Fatty acids with 6 to 12 carbon atoms, built by microbes from short acids.
What we need
Capric acid (C10), reached step by step from C6 and C8.
Division of labor
Each microbial group does one step, so the community copes with changing feed.
Why it is hard
Longer chains are toxic to the microbes, so C10 must be extracted as it forms.
Division of labor: three groups, one assembly line
  1. Residuelactose, proteins, fats
  2. Hydrolysersbreak proteins and fats into small molecules
  3. Lactate producersturn lactose and sugars into lactatee.g. lactic acid bacteria
  4. Chain elongatorsadd two carbons per cycle, from C6 up to C10e.g. Caproiciproducens, Megasphaera
  5. C6 → C10 acidsextracted in-line

Kept out: methane-forming and propionate-forming microbes, which take the same carbon and lactate away from the elongators.

The bottleneck

Every waste batch is different. That variability decides the climate benefit.

21% – 62%

Carbon-footprint reduction of a food-waste-to-SAF biorefinery using mixed-culture chain elongation. The range depends on feedstock quality and how well operating conditions are tuned to it (Zaborowska et al., 2026). Holding the consortium steady is what moves a plant to the top of that range.

Without AI 1 of 12 batches on target worst batch 47% of target output
12 whey & food-waste batches · one consortium
Lactose & sugars Proteins & fats Salts & inhibitors Lactate producers Hydrolysers Chain elongators

Illustrative model, not measured data

Feed swings

Whey and food residues change in lactose, protein, fat and salt with season, dairy product and supplier.

Elongators suffer first

The chain elongators that make caproate are the most sensitive to salts, pH and their own product.

Output collapses

When they fall behind, carbon ends up as short acids or methane instead of jet-fuel precursors.

How we unlock it

Train a digital twin in the lab, let it run production, and scale it up.

Phase 1Develop in the labon the Cellium bioreactor
  1. 01

    Set the consortium

    Hydrolysers, lactate producers and chain elongators that turn whey and food residues into medium-chain fatty acids.

    • Adaptive pressure: over successive runs we raise salt, acid levels and feed variability so only robust, productive members remain.
    • Steered selection: mildly acidic pH and short residence times keep methane-forming microbes out.
  2. 02

    Analytical tools

    Measure the feed, the products and the community itself.

    • Raman spectroscopy: in-line, every few minutes, without samples. Reads lactose, protein, fat and salt in the feed, and lactate and caproate in the broth.
    • Liquid chromatography: lab reference measurements of sugars and acids to calibrate the Raman models.
    • DNA sequencing: tracks how much of each population is in the community.
  3. 03

    Train the digital twin

    Deliberately varied lab runs on the Cellium bioreactor. 50+ signals plus the analytical data teach the twin how feed, community and output connect.

    • Planned experiments: the twin picks the next run that teaches it the most, so fewer runs are needed.
  4. 04

    Close the control loop

    The twin drives the fermenter: feed rate, pH and product removal, tested until the consortium holds under any batch.

Phase 2Run in productionevery batch, automatically
  1. 05

    Food waste batch arrives

    Whey and food-processing residues, pretreated into a liquid feed. A Raman scan gives its lactose, protein, fat and salt content before it enters the fermenter.

  2. 06

    Twin predicts and adapts

    It simulates the consortium's response hours ahead and sets the feed and conditions before problems start.

  3. 07

    Stable capric acid (C10) output

    The elongators stay active and C10 is extracted as it forms, then upgraded into SAF.

every production batch flows back and retrains the twin
Phase 3Scale upfrom laboratory to commercial conditions
  1. Laboratory

    Scale up in the laboratory

    Caproate from real residues, under first closed-loop control.

    • Lab · Cellium bioreactor
    • TUM Venture Labs
    • Trained digital twin
  2. Pilot

    Integrate and stress-test

    One process from feed to recovered caproate, run on varied batches.

    • Pilot fermenters
    • Techmi Group · TUM pilot plant
    • Robust, transferable control
  3. Production

    Validate commercially

    Production-like conditions and a business case that holds.

    • Production-representative
    • Techmi Group · SynapBioTech
    • Customer samples · cost and life-cycle analysis
AI and machine-learning models inside the digital twin
Spectral soft sensors
Partial least squares regression and neural networks convert Raman spectra into live concentrations.
Sense
Hybrid consortium model
Growth equations for each population, with a neural network learning what the equations miss.
Predict
Bayesian optimisation
Chooses the most informative next lab run and estimates how uncertain each prediction is.
Learn
Model predictive control
Plans feed and pH hours ahead from the twin's forecast, refined over time by reinforcement learning.
Act
Market pipeline

Enter aviation through high-value lubricants, then scale into fuel.

high low value per tonne small large market volume Aviation lubricants turbine & hydraulic oils Specialty esters oils, solvents, additives Biofuels & SAF jet fuel blends
Bubble size shows relative market volume · direction of travel as the process matures
  1. Enter

    Aviation lubricants

    Caproic acid is one of the C6–C10 acids esterified into synthetic turbine oil base stocks.

    $1.28Baviation lubricants market 2025, growing ~6% a year
    From laboratory and pilot output
  2. Grow

    Specialty esters & oils

    Bio-based oils, solvents and fuel additives for aviation and industry, sold in larger batches.

    Midvolume and value, steady revenue while scaling
    From pilot plant
  3. Scale

    Biofuels & SAF

    Medium-chain acids upgraded into jet-range hydrocarbons for blending.

    39.3 MtEU jet fuel demand 2025, with SAF at ~€1,925/t
    From production scale
Team

Bioprocess, sustainability and AI under one roof. 100% TUM.

Process & consortia
PG
Pablo Gómez RossiCEO & Founder · process design, lab experimentation, digital twin with hardware testing
Master's, Chemical Biotechnology, TUMInventor, EU patent applicationHardware, electronics, product design
Business & sustainability
MG
Merve GezenLife-cycle assessment, raw-material treatment, process support
Master's, Chemical Biotechnology, TUMTUM Circular Economy chairBiotech startup experience
RS
Rawan SafadiSustainability impact, techno-economic analysis, business development
Control systems & AI
TR
Thisura RanasingheAI modelling and training, data analysis
SM
Subhassh MahenthrenControl systems and hardware development
Mechanical engineerMechatronicsMaster's student, TUM
Advisors
Prof. Dr. Michael ZavrelBioprocess Engineering, TUM Campus Straubing
Prof. Fröhling
Hernan ImhoffFounder & CEO, Techmi Group
Alvaro
Gelda BogranBioprocess scale-up, Chair of Bioprocess Engineering
Partners

From lab bench to industrial vessel.

TUM Venture Labs Biotech

Munich, Germany

The Technical University of Munich's biotech startup hub. Cellium has been part of it for over two years and uses its wet lab, pilot equipment and prototyping workshop.

Techmi Group

Torrent, Valencia, Spain

Industrial process equipment manufacturer with a biotechnology line covering lab, pilot and production fermenters. Supports Cellium with mentorship, equipment and scale-up.

SynapBioTech

Valencia, Spain

Research and technology-transfer company bridging lab and factory in bioprocesses and environmental biotechnology. Supports microbial and process development.

Sources
  1. EASA, EU SAF supply exceeds 2025 ReFuelEU target (2026): 1.1 Mt SAF, 2.8% of 39.3 Mt.
  2. Fastmarkets on the EASA ReFuelEU Annual Technical Report 2026: feedstock origin, prices, 2030 capacity.
  3. Carbon Gap, ReFuelEU Aviation tracker: mandate trajectory.
  4. Transport & Environment via Skift (2025): 187.6 Mt CO₂ in 2024.
  5. Umweltbundesamt, Lebensmittelabfälle: 10.9 Mt in 2023.
  6. Astute Analytica via GlobeNewswire (2026): aviation lubricants market.
  7. Aircraft turbine oil base stock patent (US 2018/0179463): C6–C10 acid polyol esters.
  8. LfL Bayern, Facts and Figures on the Bavarian Dairy Sector in 2025.
  9. de Almeida et al., Waste-to-resource caproic acid production by anaerobic mixed microbial cultures, Biomass and Bioenergy (2025).
  10. Zaborowska et al., SAF from food waste via medium-chain carboxylates, Elsevier (2026).