Verify and Validate: Using ChatGPT to Tame Suya Pepper for Cooking and Eating
There is an old saying, “I can’t trust you, further than I can throw you.” This saying applies to generative AI in the sense that I find it hard to trust it beyond my ability to verify and validate its responses. Fortunately, there are many use cases amenable to “human in the loop” success. One use case I am increasingly enjoying is modifying and creating recipes. While my conversations with ChatGPT in the kitchen have taught me some fascinating elements of food science and chemistry, a taste test at the end of the experience is the ultimate opportunity to verify and validate.
I was particularly appreciative when I embarked on a mission to transform suya pepper that I bought in Nigeria from atomic hot to tolerably hot.
The First Prompt: Starting Broad
I began the interaction with the following prompt:
“I have a LOT of suya pepper that is far too hot and spicy to cook without some dilution. Please recommend a recipe for diluting the suya spice while preserving the flavor.”
ChatGPT responded with a blend that included roasted peanuts, sweet paprika, and sugar—sensible ingredients for softening heat. But I had dietary goals and pantry realities constraining my cooking and eating, so I continued:
“This is a good start. Let’s revise so that the recipe minimizes additional sodium, does not add any more heat, and removes peanuts.”
ChatGPT quickly responded, removing the peanuts, omitting additional chili, and dialing back the salt. However, I noticed a pattern—when I offered a list of additional ingredients available in my pantry, ChatGPT included all of them rather than selecting with discretion. That generated a refining prompt from me:
“Wait. Why did you add all the spices I mentioned? I also provided those as options. Please revise the recipe according to a precise blend that takes into account what I do not have.”
This moment was key. ChatGPT shifted from guessing to anchoring. It asked nothing, but responded with a more grounded approach: no sweet paprika or coriander (which I lacked), and instead a balanced mix based on smoked paprika, turmeric, nutmeg, and mild curry powder—ingredients I confirmed I had on hand.
The Final Diluted Suya Spice Blend
We arrived at this custom recipe:
Diluted Suya Spice Blend (Based on My Pantry)
- 4 tablespoons original suya pepper (the hot stuff)
- 1½ tablespoons smoked paprika
- 1 tablespoon onion powder
- 1 tablespoon garlic powder
- 1 tablespoon ground ginger
- ½ teaspoon turmeric
- ¼ teaspoon nutmeg
- ½ teaspoon mild curry powder (no chili)
- 1 teaspoon coconut sugar or brown sugar (optional)
- Salt omitted by default
Instructions
- Combine all ingredients in a bowl and mix thoroughly.
- Taste test a small pinch. If it’s still too hot:
- Add 1 additional tablespoon of onion powder or smoked paprika, mix again, and retest.
- Store in an airtight jar away from heat or moisture. Shake before each use.
Why These Ingredients Work
- Smoked paprika substitutes for sweet paprika with smoky depth.
- Turmeric + nutmeg complement the savory profile and soften heat perception.
- No coriander or sweet paprika were used, since I did not have them.
- No peanuts or salt were added by default, to meet my dietary constraints.
Applying the Blend: Suya-Sautéed Shrimp
Once I had a usable blend, I prompted:
“Using this mix, give me a revised recipe for suya sauteed shrimp using 20 Master Catch raw shrimp.”
I included the brand name of the shrimp in case that information gave ChatGPT important information on size, texture, or cookability of the shrimp. I next prompted::
“I don’t have lemon or lime juice. Is there an alternative recipe you can recommend?”
With that information, ChatGPT produced a flexible shrimp recipe that used apple cider vinegar in place of citrus and stayed true to the seasoning profile I had developed.
Suya-Sautéed Shrimp (Low Sodium, No Added Heat)
Serves 2–3
Ingredients
- 20 raw Master Catch shrimp, peeled and deveined
- 1 tablespoon diluted suya spice blend (above)
- 1 tablespoon olive oil
- 1 teaspoon apple cider vinegar (or ½ teaspoon mild mustard if no vinegar)
- ½ teaspoon coconut sugar (optional, balances bitterness)
- Optional garnish: chopped parsley or cucumber slaw
Instructions
- Prep: Pat shrimp dry with paper towels.
- Marinate (10–15 minutes):
Combine shrimp, suya spice blend, vinegar (or mustard), and optional sugar in a bowl. Toss thoroughly and let sit at room temperature. - Sauté:
- Heat oil in a skillet over medium-high heat.
- Add shrimp in a single layer and cook 2–3 minutes per side, until pink, opaque, and lightly browned.
- Serve Immediately:
- Garnish as desired.
- Serve with plain rice, roasted plantains, or a yogurt-cucumber dip to cool the heat further.
Verify and Validate
I tasted one shrimp first and found near perfection. I was able to tolerate the heat, and the full flavor of the suya came through loud and clear. Next, I warmed up a slice of sweet potato pie as a smooth complement; it cut the heat on my tongue with a bonus burst of sweet and savory. My “verify and validate” process was a success. I recorded the recipe for future use.
Prompting Lessons from the Kitchen
This session clarified some broader lessons for working with ChatGPT—especially in use cases where verify and validate is not just desirable, but necessary:
- Prompt iteratively, not magically. Your first request is rarely the final recipe. Start wide, then refine with specificity.
- List what you do and do not have. Ingredient lists matter. AI cannot infer context, you must provide context for improved results.
- Push for grounded responses. If AI is guessing, correct the behavior. Ask it to acknowledge constraints.
- Use your senses. Ultimately, cooking is physical. Taste, texture, aroma—they are all tools for post-prompt verification and validation.
- Keep the loop tight. Generative AI is not your chef—it is your assistant. You still need to wear the apron and wield the measuring spoons in the kitchen.
Final Thoughts
I did not just create a customized recipe—I experienced a collaboration. The suya pepper blend I can now use is one I could not have invented on my own. I relied on an iterative conversation and tools to verify and validate along the way. My taste buds and the sweat glands in my forehead can verify that this version of suya shrimp is a keeper for me.
If you find yourself wrestling with a challenging ingredient or a dietary constraint, consider inviting ChatGPT into the kitchen for a collaboration session. Just make sure you test and taste along the way to your plate.


I read a few of your Ahan Analytics articles and looked at 1-twenty two. You have some great sites.
Per this AI article, I loved the testing in the kitchen approach. I appreciate and have on my own, “…experienced a collaboration…” I look at these AI agents as Digital Assistants.
Over the past six months, I’ve been utilizing my company’s paid subscription to ChatGPT. I often upload source documents for the tool to “LEARN”. I make those files the constraints as I want, “grounded responses” and not “AI is guessing” (which I find worthless and can invalidate the whole session). I do make my sessions interactive. To start, I try to include basic instructions such as (Note: while I have a couple of Masters degrees, I find the responses too wordy, so, I’ve been opting for ’12th grade reading level’ and usually with decent results):
ONLY reference the uploaded files and text in all answers
QUESTIONS TO RESEARCH AND ANSWER:
Briefly evaluate this tax based on all uploaded files AND the 10 tax revenue characteristics
STYLE & FORMATTING INSTRUCTIONS:
– 2 paragraphs max answer
– create text at a 12th grade reading level
– detail file reference sources
and, sometimes instead of files, I provide it as much non-edited, my bullets and outline as possible…
ChatGPT: create a Job Description for “Oracle Retail Revenue Settlement & Reconciliation Techno-Functional Senior Consultant” using the below outline
STYLE & FORMATTING INSTRUCTIONS:
– create text at a 12th grade reading level
– concise, American technical, professional
– use bullets
SUMMARY OF JOB DESCRIPTION:
Job Title: Oracle Retail Revenue Settlement & Reconciliation Techno-Functional Senior Consultant Lead
Location: Pleasanton CA USA or Coppell TX USA, (no remote. no relocation)
Role Overview
– Primary Oracle Retail Revenue Settlement & Reconciliation Techno-Functional Senior Consultant, all integrations to Oracle Retail and Oracle Fusion
– operations and Oracle Retails applications
– Leadership & Team Management – onsite and offshore Oracle Retail development and support teams
– Communication, Stakeholder Management & Cross-functional collaboration
Key Responsibilities
Required Skills, Qualifications
Experience, 10 years, at least 2 full lifecycle implementations, at least leading 2 support teams
Oracle Retail Reference Model (RRM) and implementation method, Retail Business Process Flows, Retail Technical Models
Application [chatgpt: expand with lots of module function and features] Oracle Fusion Financials Functional Business Analyst
Technical
Oracle Retail desktop data upload tools (Foundation Data Load, Spreadsheet Data Loader, batch loaders)
Oracle Retail reporting tools (Operational Insights, dashboards and in-context, Oracle Retail Insights, Retail Science Platform, Merchandising BI Pub)
Preferred Skills
enter all Oracle Retail modules
Systems Integration Tools including RICS, RIB, RSB, BDI, RFI, EICS (older mapping RXM, RBO, SIM, RWMS, RPM, EXT, AIP), POS integrations, api, jms
Security – Retail Apps Admin Console (RAAC), define security roles, assign, admin access