Three real runs. Unedited.

 

IntelliForm in Practice: Three Real Runs, Unedited
IntelliForm v0.6 · Showcase Run · April 2026

Three real runs.
Unedited.

After the parser and controller reliability upgrade, we captured three IntelliForm runs exactly as they came out — two successes and one honest infeasibility. Here they are, with full blend compositions, costs, eco scores, and regulatory output.

Platform  intelliform.streamlit.app Repo  github.com/Cheme-Nova/IntelliForm Captured  2026-04-24 v0.6 · Post-Upgrade

Demos are easy to cherry-pick. What I want to show here is the actual output of the upgraded IntelliForm controller — three runs chosen to represent the range of what the platform does today: a clean agricultural success, a food-vertical success with real regulatory flags, and a case where the optimizer correctly refused to produce a result rather than hallucinate one. All three were captured locally on April 24, 2026, after the parser, vertical-mapping, and constraint-propagation upgrades. Nothing has been edited.

Runtime note: These runs used LLM_PROVIDER=regex — the local regex-based parser rather than a live LLM endpoint. This is intentional: it isolates the optimizer's performance from any LLM variance, demonstrating that the constraint extraction, ingredient filtering, and optimization stack work correctly independent of the language model backend.

Run 1: OMRI Bioinsecticide Adjuvant

Agricultural vertical. The brief asked for an OMRI-listed bioinsecticide adjuvant for organic crops, under $8/kg. This is a constrained problem: organic compliance limits the available ingredient pool, and adjuvant performance requirements are specific.

Agricultural · Run 1 of 3
OMRI Bioinsecticide Adjuvant for Organic Crops
✓ Optimal
BRIEF › OMRI-listed bioinsecticide adjuvant for organic crops, under $8/kg
$1.62
Cost / kg
82.9%
Bio-based
78.2
Perf Score
87.7
Eco Score
A
Eco Grade
132
Ingredient Pool
Optimized Blend Composition
Canola Methyl Ester
65.1%
Ethanol (agri grade)
20.2%
Sulfur Dust (agri)
9.5%
Urea (agri)
1.0%
Ammonium Sulfate (fertilizer)
1.0%
Calcium Ammonium Nitrate
1.0%
Regulatory Output
EPA FIFRA · EU Reg 1107/2009 · OMRI Listed · NOP (USDA Organic)
Ethanol (agri grade): OMRI Listed
Sulfur Dust (agri): OMRI Listed
No EU-banned active substances detected

What this shows: The optimizer resolved the agricultural vertical correctly, filtered to a 132-ingredient pool, and returned a feasible blend at $1.62/kg — well under the $8/kg ceiling. The eco score of 87.7 (Grade A) reflects the high canola methyl ester content, a bio-derived carrier. Both ethanol and sulfur dust came back OMRI-flagged as listed.

Still needs: Tank-mix compatibility and field performance require agronomy validation. Some ingredients need manual OMRI verification rather than assumed organic acceptance. This is a first-pass formulation direction, not a registration-ready product.

Run 2: Clean-Label Plant Milk Emulsifier

Food vertical. The brief asked for a clean-label emulsifier blend for plant-based milk, GRAS compliant, under $4/kg. The food vertical carries stricter allergen obligations than any other, and the optimizer surfaced them correctly rather than suppressing them.

Food · Run 2 of 3
Clean-Label Emulsifier for Plant-Based Milk
⚠ Review Required
BRIEF › Clean-label emulsifier blend for plant-based milk, GRAS, under $4/kg
$3.39
Cost / kg
90.9%
Bio-based
71.9
Perf Score
83.4
Eco Score
A
Eco Grade
217
Ingredient Pool
Optimized Blend Composition
Water for Injection (WFI)
53.9%
Lauryl Glucoside (and) Glycerol
25.1%
Sodium Bicarbonate (agri)
3.7%
Resistant Starch · Psyllium Husk · Pea Fiber · Oat Fiber · Bamboo Fiber
1.0% ea.
Pea Protein · Soy Protein · Whey Protein · Casein · Egg Albumen · Transglutaminase · Chlorella · Spirulina
1.0% ea.
Regulatory Output
FDA GRAS · EU Food Additives Reg 1333/2008 · EFSA · Codex Alimentarius
Clean label compliant
No EU-banned food additives detected
⚠️ Soy Protein Isolate: Allergen declaration required (soy)
⚠️ Whey Protein Isolate: Allergen declaration required (dairy)
⚠️ Casein (food): Allergen declaration required (dairy)
⚠️ Egg Albumen Powder: Allergen declaration required (egg)
🚩 MANDATORY ALLERGEN LABELING: Soy, Dairy (whey/casein), Egg

What this shows: The optimizer found a feasible food-vertical blend at $3.39/kg against a $4/kg ceiling — clean label compliant, 90.9% bio-based, eco grade A. But it also surfaced four mandatory allergen declarations without being prompted to look for them. A generic LLM would either miss these or note them as suggestions. IntelliForm flags them as mandatory, citing the specific regulatory framework (EU 1333/2008, Codex Alimentarius).

The performance score of 71.9 — slightly below the 75.0 constraint — is worth noting. The optimizer accepted this within tolerance rather than producing a false pass. The "Review Required" status reflects real gaps, not system failure.

Still needs: The current optimizer is better at finding feasible ingredient sets than fully optimized food architecture. This formulation is a first-pass ingredient direction — useful for R&D scoping, not a finished commercial beverage system. The allergen flags are real and require action before any commercial use.

Run 3: The honest failure

Industrial vertical. Low-VOC degreaser for heavy equipment — high flash point, strong grease lift, moderate foam, bio-based, under $4/kg. This is where IntelliForm does something most AI formulation tools won't: it refuses to produce a result.

Industrial · Run 3 of 3
Low-VOC Industrial Degreaser for Heavy Equipment
✗ Infeasible
BRIEF › Low-VOC industrial degreaser for heavy equipment with high flash point, strong grease lift, and moderate foam.
Cost / kg
Relaxation Rounds
158
Ingredient Pool
No SVHCs
❌ Swarm halted — Optimization failed
Could not find feasible blend after 3 relaxation round(s).
  • Increase max cost above $4/kg — high-flash-point bio solvents are inherently more expensive
  • Reduce the bio-based floor below 80% to allow high-performance synthetic carriers
  • Adjust max ingredient % to allow a dominant carrier solvent
  • Switch to All Verticals mode to access cross-vertical high-flash solvents
Regulatory Output (pre-blend screen)
REACH · EPA Safer Choice · NSF/ANSI 61 · NSF A1 (food-safe)
No SVHC substances detected in the filtered industrial pool

Why this matters more than the two successes: The constraints here are genuinely in conflict. Bio-based solvents with high flash points and strong grease lift at under $4/kg do not exist in sufficient concentration within the current industrial ingredient pool to satisfy all three objectives simultaneously. After three relaxation rounds — each progressively loosening the constraints — the optimizer still found no feasible blend and halted.

An AI that halts and explains why is more useful than one that produces a confident answer to an impossible problem.

The output includes specific, actionable guidance on which constraint to relax and why — not a generic error message. It also completed the regulatory pre-screen on the pool itself, correctly returning no SVHC hits even though no blend was produced. The system continued to do useful work up to the point of infeasibility.

What this tells us about the roadmap: Industrial cleaning needs deeper formulation architecture — specifically, better ingredient-package logic for high-flash-point bio solvents and a richer industrial pool. This is a known gap and an active development target, not a surprise. Showing it openly is more credible than hiding it.

What these three runs actually prove

🎯

Constraints drive the optimizer

Cost ceilings, bio-based floors, and performance targets parsed from plain English actually propagate into the optimization — the $1.62/kg result against an $8/kg ceiling is not coincidence.

🏷️

Vertical isolation works

Agricultural, food, and industrial runs each drew from separate ingredient pools (132, 217, 158 ingredients respectively) with no cross-vertical contamination.

⚠️

Regulatory flags are real

The four allergen mandatory declarations in Run 2 were not prompted — they emerged from the regulatory layer against EU 1333/2008 and Codex Alimentarius automatically.

🚫

Honest infeasibility beats hallucination

Run 3 attempted three relaxation rounds before halting. It returned a diagnosis and four specific remediation paths — not a fabricated blend.

🌱

Eco scoring is independent

Both successful runs returned Grade A eco scores (87.7 and 83.4) computed independently of the optimization objective — a separate accounting, not a reward metric.

🔬

No LLM required for core function

All three runs used the regex parser backend. The optimization stack — constraint extraction, filtering, solving — runs fully locally, independent of any LLM provider.


Try it yourself

The platform is live at intelliform.streamlit.app. No account, no credit card. Write a formulation brief in plain English — application, cost target, bio-based requirement, certification needs — and see what the optimizer returns. The full source is MIT-licensed at github.com/Cheme-Nova/IntelliForm.

If your formulation needs move toward a physical pilot batch, ChemRich Global runs 50–200 kg pilot manufacturing from our New Jersey facility, with a no-cost reformulation guarantee if certification criteria aren't met. Contact: shehan@chemenova.com

Comments

Popular posts from this blog

We Built an AI Formulation Co-Pilot for the Specialty Chemicals Industry. Try It Free.

Custom Manufacturing 2.0: Navigating the 2026 Agentic AI Revolution

Beyond Predictive Modeling: The Rise of the Agentic Chemical OS