Early Stage Investment Opportunity
POMA AI - The Context Engine
Every company's competitive advantage lives in their knowledge—contracts, policies, customer data, domain expertise. Getting that knowledge into agentic AI context—properly and at scale—is the problem everyone talks about. Poor ingestion and chunking waste up to 99% of computational power while agents hallucinate. POMA AI solves this, enabling intelligent knowledge orchestration for agentic systems.
Funding Round: Total 1.5 Mio € CLA, 500k € left
Convertible Terms: 8 Mio € Cap, 25% Discount
RAG Pipeline Economics: Where Quality Is Won or Lost
Retrieval Augmented Generation (RAG) looks simple on paper — ingest documents, retrieve relevant passages, then answer with correct context. In practice, the quality ceiling is set much earlier, at the chunking layer. Most failures are system design problems, not model IQ problems.
RAG Pipeline
Data Sources
PDFs, pics & video, contracts, xls, tickets, transcripts, wikis, code — the raw context that defines the knowledge boundary of your system.
Ingestion
Parse, normalize, and clean raw documents. Extraction quality here propagates directly into chunk and embedding quality downstream.
Chunking
The highest-leverage decision in the pipeline. Determines what each vector "means" and what evidence can ever be found at query time.
Vector Store and Retrieval
Ranks chunks by similarity — often with hybrid search and reranking — before passing raw context to the LLM for synthesis. Bad retrieval = hallucination risk.
Notable Companies
Sharepoint, Google Drive, Confluence

Llamaparse (LlamaIndex), Unstructured.io, Mistral OCR, POMA AI

POMA AI. No otherplayers with proprietary tech, naive linear chunking only

MongoDB, Pinecone, qdrant, weaviate

The reality: enterprise RAG pipelines break on hard queries while token waste explodes
The Core Tradeoff
Precision vs. context continuity. Smaller chunks improve retrieval precision but lose surrounding context. Larger chunks preserve continuity but introduce noise that degrades similarity matching.
Why Chunking Is the Quality Gate
Every downstream component — the retriever, reranker, and LLM synthesizer — operates on whatever evidence chunking surfaces. Bad splits produce confident but wrong answers. The retriever can't fix what was never retrievable.

Naive Chunking Kills Context And Creates Orphans
The Model Doesn't Say "I Don't Know" - It Hallucinates
Getting It to Work Costs Too Much Time and Money

In 2026 alone, enterprises are projected to waste over $10 billion (and its equivalents in wasted energy and avoidable CO₂ emissions) annually on inefficient document processing methods.
As of April 2026, the top individual user on Meta's internal 'Claudeonomics' leaderboard consumed 281 billion AI tokens in a 30-day period
"Compute costs are more expensive for us than a lot of other things."
— Winston Weinberg, Founder & CEO, Harvey
The POMA AI Context Engine Solves This
One API Call. The Entire Retrieval Stack.

POMA AI is a fully managed Retrieval-as-a-Service platform (RaaS). A single API call replaces Unstructured, Pinecone, and all the custom glue code in between. POMA AI owns the full retrieval pipeline end-to-end.
Key differentiator: Pipelines based on POMA AI deliver more relevant context and have a lower total cost of ownership — not marginally, but structurally so.
1
Document Ingestion
Any format, at scale, with free ingestion on all paid tiers.
2
Intelligent Chunking
Proprietary and patented, context-preserving chunking. No meaning lost at boundaries.
3
Embedding + Storage
Managed vector storage. No infrastructure to provision or maintain.
4
Precise Retrieval
Accuracy-per-token: fewer tokens needed, lower cost, correct answers.
Superior Recall Per Token Proven
In our officeqa benchmark, traditional solutions either do not find the answer due to context window overflow or require more than 4x the token budget. Only POMA AI does and delivers high accuracy per token.
Note: for some questions, naive chunking and unstructured.io require a token budget above 1 mio tokens, which in production leads to context window overflow, resulting in the correct context not being found.
The Stakes Are Too High for Context to Fall Short
In legal, finance, healthcare, and regulatory a wrong answer isn't an inconvenience. It's a liability. These are the sectors where retrieval accuracy has a price tag and it's exactly why many systems are stuck in POC.
$10.4B Legal AI market by 2035
28.3% CAGR. LLM-based drafting, automated compliance, and AI-assisted litigation tools.
$10.3B AI in finance market by 2029
38.1% CAGR. Data analysis, automation, KYC, AML, risk reporting, loan files, insurance claims.
$13.9 B clinical documentation market by 2030
28.3% CAGR. AI-driven documentation systems and retrieval from clinical records, EHRs, insurance documents.
$12.3B RegTech by 2030
36.9% CAGR. Compliance management, fraud detection, regulatory radars.
POMA AI's Addressable Slice — Bottom-Up
  • SAM: ~9,000–16,000 EU SMEs/scale-ups in target verticals with document-heavy AI use cases
  • Enterprise upside: the large enterprise market unlocks via a sales-led motion
What Retrieval Failure Costs in These Verticals
  • Legal: A missed precedent or misquoted clause — €50K–€500K in malpractice exposure per case
  • Finance: An AML false negative — €1M–€50M in regulatory fines (EU AMLD)
  • Healthcare: A wrong clinical document retrieval — patient safety risk, GDPR liability
  • Regulatory: An inaccurate compliance report — license suspension, reputational damage
  • Plus: hundreds of man days to duct tape a leaky pipeline
At €0.01–0.02/query, POMA AI's cost is trivial versus the downside it prevents.
Two Products. One Platform. Every Entry Point.
POMA AI meets customers where they are — whether they're building from scratch, augmenting an existing pipeline, or just need better document processing.
POMA Grill: Context Engine (RaaS)
Fully managed retrieval pipeline starting at €0.01/query. Replaces the entire in-house ingestion, storage and retrieval stack. Per-query pricing with monthly tiers that auto-incentivize upgrades.
POMA PrimeCut: Ingestion+Chunking Only
Intelligent extraction and chunking starting at €0.003/page, pay-as-you-go. For teams that want POMA AI's chunking quality without touching their storage and retrieval layers. Immediate value, zero commitment.
RaaS - Simple Usage-Based Pricing
Free ingestion on all paid tiers. No hidden infrastructure fees. Upgrade economics are automatic — customers save money by moving up the tier ladder.
Upgrade Triggers: Every Transition Is Economically Obvious
No sales call required for the first three tier transitions. The math does the selling — customers save money by upgrading at each threshold.
75–96% Gross Margins Across All Paid Tiers
Smart infrastructure design keeps COGS structurally low and margins attractive.
Infrastructure & COGS: Built for Margin
Architecture Decisions That Drive Margin
  • No reranker in the stack — eliminates the largest variable cost component in competing RAG architectures.
  • Multi-tenant by default — 10 customers sharing infrastructure keeps per-customer COGS at €16–67/month.
  • No Unstructured dependency — proprietary extraction eliminates third-party per-page fees.
  • Free ingestion cost runs ~€300 per 100K pages, absorbed as customer acquisition cost.
COGS Breakdown per Customer/Month
Ingestion: €0.003/page — fully absorbed by POMA across all paid tiers, treated as a CAC investment and amortized over CLT.
Go-to-Market: PLG First, Sales-Led at Scale
POMA AI's GTM is engineered so that upgrade economics are automatic. Customers don't need convincing — the math does it for them. Developer adoption at the free tier seeds the commercial pipeline.
Phase 1 — Product-Led Growth (Now)
Developer self-serve. No credit card required for the free tier. Sign up in 2 minutes, integrate in 15. Free → PAYG → Pro upgrades are economically automatic.
  • Target: 100 free signups/month
  • 8% Free-to-PAYG conversion
  • 25% PAYG-to-Pro over customer lifecycle
Phase 2 — Sales-Led (€299+ MRR)
Business and Enterprise tiers are sales-assisted. Annual contracts with 10% discount. Enterprise = full consolidation sale, replacing 2–4 in-house engineers.
  • Enterprise tech budget savings: €185K–525K/year per customer
  • Annual contracts create revenue predictability and reduce churn
Selected Early Customers — Live in Production
ki-ra.eu
RAG for Regulated SME
Talk to your data
ki-ra builds RAG pipelines for small to medium sized law firms, insurance brokers and other regulated players in a secure cloud environment allowing them to harvest their data for business intelligence.
20,000+ pages ingested/month
HeyData
Compliance Automation — Regulatory Vertical · heydata.eu
Regulatory document retrieval for 2,000+ SME compliance customers
HeyData's compliance platform covers GDPR, NIS2, and ISO 27001 for SMEs. POMA AI ingests 3,000–5,000 pages/month of regulatory frameworks, audit documentation, and legal guidelines — powering accurate compliance Q&A across HeyData's customer base.
3,000–5,000 pages ingested/month · Regulatory vertical
Heuristiq — Channel Partner
Channel Multiplier · RAG for Regulated Industries · heuristiq.co
AI strategy and RAG implementation specialists for regulated industries — legal, compliance, and PE-backed companies. Heuristiq builds RAG pipelines for POMA AI's core target verticals and actively recommends POMA AI to their client base as the managed retrieval layer.
4
Early Customers
Live in production
10-20K
Pages/Month
Ingested per customer
3
Verticals
Regulatory, Finance + Sales Intelligence
2
Channel Partner
Specialist RAG advisors
Unique Competitive Positioning
Chunking is where the context pipeline is won or lost. It's the only stage no competitor has solved at production quality — except POMA AI.
🟢 = core competency
🟡 = part of offering but not core competency
🔴 = not part of offering
Competitor deep dive:
Three-Layer Competitive Moat
POMA AI's defensibility isn't a single feature — it's three compounding layers that get harder to replicate with every customer added.
Free Ingestion Lock-In
Customers pay nothing to index — but switching means re-ingesting their entire corpus plus weeks of re-validation. Data network effect.
Proprietary, Patented Chunking as Core IP
Ingestion + intelligent chunking in a single managed pipeline. Not replicable from off-the-shelf components. Improves as the ingested corpus grows.
Customer Feedback Loop
Customer feedback on retrieved context quality. Aggregated across customers, this is a proprietary training dataset no new entrant can replicate without first acquiring customers.

The moat is self-reinforcing: better chunking → more customers → more query data → better chunking and retrieval.
Our Team
Experienced team already in place - focus on execution not on hiring.
Dr. Alexander Kihm
Founder, CEO & MD
Ph.D. Big Data Econometrics at German Aerospace Center
Serial entrepreneur:
Advo Assist, fairr (exit to Raisin)
25+ yrs coding experience
Business & Sales Team
Chief Commercial Officer & MD (signed)
Serial entrepreneur, PhD, 20+ yrs in startups
Annika, Head of Marketing
10+ yrs experience marketing & growth
Florian, Business Developer
7+ yrs experience
Jens, Finance & Legal (ext)
20+ yrs experience in finance, legal, startups (exit to Raisin)
Product & Engineering Team
Ryan, Director of Product
15 yrs product experience as founder and manager
Sepehr, Gen AI Developer
10+ yrs machine learning experience
Fabia, Senior Developer
PhD, 10+ yrs coding experience
Raffael, Senior DevOps
10+ yrs backend
Suat, Senior UX Dev
10+ yrs full-stack
Alexander, Sales Dev:
5+ yrs coding experience, BSc thesis on AI+RAG
Selected Team LinkedIn Profiles
Hiring Roadmap
  • Account Executive (Q4 2026)
  • Customer Success (Q1 2027)
Funding — Capital Stack & Use of Funds
Total Capital Available — ~€2.5M+
CLA Seed Round
€1.5M · €1M committed · €500K open · Rolling CLA
Terms: €8 mio cap, 25% discount
Status: CLOSING
Public Grants & Loans
~€900K · Invest program · Expected Q3-2026
Status: IN PROCESS
Cloud Credits
>$200K · Google Cloud · AWS · Azure · Stripe · Infrastructure runway through end of 2026
Status: SECURED
Cloud credits effectively eliminate infrastructure COGS through end of 2026 and beyond, extending cash runway materially.
Product & Engineering — 55%
RaaS offering & Ingestion API · Complete retrieval pipeline · Accuracy benchmarks
Go-to-Market — 30%
Developer marketing · Content & SEM/SEO · First sales hire at Business tier · Channel partner activation (like Heuristiq)
Operations & Compliance — 15%
Legal · Finance · GDPR compliance · Runway buffer
Investors (selection)
heartfelt.vc
Berlin-based early-stage VC
b# (b-sharp)
Berlin-based deep tech VC
Till Faida
Founder, Adblock · Serial entrepreneur & angel investor
Dr. Jan Wildhirth
Partner & Global Head of AI, Fieldfisher
All investors committed within current CLA round.
€500K remaining to close · Total round: €1.5M CLA · Runway to late 2027 incl. public funding
Next Steps & Timeline
We have a clear roadmap to rapid scaling with defined KPIs and milestones to track progress and prepare for our Seed Round in 2027.
H2 2025
✓ Company founded
✓ Launched beta program with German pilot customers
✓ Launched ingestion API for PDF and 50+ file formats
✓ Hired DevOps, Gen AI and BizDev
Q1 2026
✓ First revenue milestone
✓ Published benchmark and full go-live
✓ Patent granted
✓ Hired Director Product & Head of Marketing
Q2 2026
Close Funding Round
Launch RaaS
Sign RaaS Design Partners
✓ Hired CCO

H2 2026
First RaaS revenue
Identify key initial vertical for scaling
First business tier customer

2027
Business and Enterprise tier sales-assisted motion
Seed round ready
Ground-Floor Investment Opportunity
at the AI and Energy Inflection Point
POMA AI is positioned to transform how companies leverage AI by eliminating the massive inefficiencies in current document processing systems.
With patented technology, European data sovereignty, and a clear path to commercialization, we invite you to join us at this critical inflection point in AI adoption and Decarbonization.
Contact: Dr. Alexander Kihm