Forecasta

BACKGROUND

Introduce your business and what you do there.

I’m the CEO of a company that develops advanced radar-based monitoring systems to enhance safety and efficiency in industrial spaces by detecting movement, preventing collisions, and protecting people and equipment in factories, warehouses, parking facilities, and job sites.

OPPORTUNITY / CHALLENGE

What challenge were you trying to address with Forecasta?

We needed to get rid of the manual per-sensor tuning process on our manufacturing line hardware. It worked when freshly installed, but the settings would go stale silently the first time a line changed. There was no reliable way to tell hardware people the sensor had drifted until it started behaving badly.

SOLUTION

What was the scope of their involvement?

We’ve hired Forecasta to build the machine learning layer that reads sensor data and makes the judgment call itself. They’ve also been tasked with making the calibration process self-monitoring so it flags drift instead of hiding it.
Forecasta has built a classification model that reads temporal patterns in sensor data, approach speed, how reflected energy changes, how long something lingers, and whether motion is cyclical. This model sorts what a sensor sees into five categories: a person, a vehicle or piece of equipment, routine machine motion, a static structure, or noise.
On top of that, Forecasta has built the self-calibration layer that runs automatically at installation and continues to check its own accuracy afterward. The whole system has been evaluated using line-disjoint testing, meaning it’s been validated on production lines that were never part of the training. This is a stricter bar than testing on more data from the same lines we started with.

What is the team composition?

We work with 2–5 teammates from Forecasta.

How did you come to work with Forecasta?

We found them through a referral. We chose them over other options because they had high ratings, were close to my geographic location, had pricing that fit our budget, and were referred to me.

What is the status of this engagement?

We started working with them in March 2026, and the engagement is ongoing.

RESULTS & FEEDBACK

What evidence can you share that demonstrates the impact of the engagement?

Calibration and classification have been automated. Our per-sensor setup process has changed from something a technician had to hand-tune that would quietly degrade over time to something that calibrates itself and tells us when it’s drifted. We didn’t formally track numbers like nuisance alert rate or exact commissioning time before and after, which in hindsight, we wish we had.
However, I can say confidently that the operational change from manual to unattended was real and immediate at every install after this was shipped. We’ve also reduced the manufacturing line manual work. The test setup has been automated, saving 5–10 minutes per unit.

How did Forecasta perform from a project management standpoint?

They’ve delivered on time. We communicate via in-person and virtual meetings.

What did you find most impressive about them?

They have robust and creative project management. I’m impressed by their outside-the-box thinking. They understand the problem and come up with minimalist solutions for low-resource embedded systems with a low budget.

Are there any areas they could improve?

There’s nothing they can improve on.

BACKGROUND

Please describe your company and position.

I am the Founder of Faberiq.ai

Describe what your company does in a single sentence.

Faberiq.ai is an AI front office for small service businesses. It handles customer conversations across Instagram, WhatsApp, Facebook, SMS, email, and web chat, then qualifies leads, books appointments, sends quotes, and collects payments. Businesses connect their own tools, including Google and Microsoft calendars, Square, and Stripe, and can build custom automations in a visual workflow designer. Agents answer from the business's own documents through a knowledge base, so responses are grounded in what the business actually offers. We also serve agencies that white-label the platform for their clients, with a branded customer portal and embeddable widgets.

OPPORTUNITY / CHALLENGE

What specific goals or objectives did you hire Forecasta to accomplish?

• Make the platform's citation guarantee hold up in production across messy, real world documents such as scans, exports, tables, and nested clauses, not just clean text.
• Preserve verifiable provenance (document, page, exact position) through every stage of the pipeline, so every citation resolves to a real, checkable location.
• Build a calibrated "I don't know" path, so the agent says so instead of answering from general knowledge when retrieval finds nothing relevant.

SOLUTION

How did you find Forecasta?

Referral

Why did you select Forecasta over others?

• High ratings
• Great culture fit
• Referred to me
• Company values aligned

How many teammates from Forecasta were assigned to this project?

1 Employee

Describe the scope of work in detail. Please include a summary of key deliverables.

Forecasta owned one subsystem inside our platform: the pipeline between a customer uploading a document and our agent answering with a verifiable citation. They built it to our existing interfaces and worked alongside our internal team, who kept ownership of the workflow engine, agents, integrations, and everything else. The scope covered four stages. First, parsing documents in whatever format customers actually upload, including OCR for scans, and recording the source document, page, and position of every extracted piece of text. Second, chunking along each document's own structure, so tables stay intact and clauses stay whole instead of being cut at a fixed length. Third, hybrid search that combines keyword and semantic matching, so paraphrased questions find the right passage and exact terms like identifiers or numbers aren't missed. Fourth, returning answers with citations that resolve to the exact source location, and returning an explicit "insufficient grounding" response when nothing relevant is retrieved. The key deliverables were the production pipeline itself, the abstention behavior, an evaluation set and harness that measures citation accuracy so we can test any pipeline change against a number, and handover documentation for our internal team.

RESULTS & FEEDBACK

What were the measurable outcomes from the project that demonstrate progress or success?

Citation accuracy on a held out evaluation set (whether a cited span actually supports the claim) became measurable for the first time, and is now scored independently from answer quality. Add specific accuracy or unsupported answer rate figures here if available.

Describe their project management. Did they deliver items on time? How did they respond to your needs?

It was a fixed-scope engagement built against our interfaces, and communication went directly engineer to engineer the whole way through, with no account management layer in between. That kept decisions fast and meant questions got answered by the people actually writing the code. They were responsive to our requirements and stayed disciplined about scope, building their subsystem without drifting into the rest of our platform. 

When their first chunking approach started breaking tables and clause structure, they caught it themselves from the evaluation numbers and rebuilt that stage, about a week of work, rather than shipping something that scored fine on average but failed on the content that matters most. The parsing layer took longer than originally scoped because both sides underestimated how many document formats our customers upload, but they flagged it early and [the project still wrapped in April as planned

What was your primary form of communication with Forecasta?

• Virtual Meeting
• Email or Messaging App

What did you find most impressive or unique about this company?

They treated our citation requirement as the actual spec, not a nice-to-have. Plenty of teams can stand up a retrieval system that looks good in a demo. Forecasta took "every claim must point to a verifiable source" and let it drive every design decision, from how documents are parsed to how they're chunked. The clearest example was their chunking rebuild: average retrieval scores looked acceptable, but they dug into the evaluation results, saw citations failing specifically on tables and clauses, and fixed it even though nobody would have noticed from the headline numbers. That willingness to hold themselves to the hard requirement, backed by real measurement, is what set them apart.

Are there any areas for improvement or something Forecasta could have done differently?

Two things. First, the range of document formats our customers upload was underestimated at the start, on both sides, and the parsing layer took longer than scoped as a result. A short discovery phase sampling real customer files before committing to the plan would have caught that earlier. Second, handover documentation was mostly produced at the end. Writing it alongside the build would have let our internal team ramp up sooner and review decisions as they were made.

BACKGROUND

Introduce your business and what you do there.

I’m the CEO and co-founder of UtterVision, an automated platform that simplifies and streamlines visa and immigration applications for multiple countries. We offer step-by-step guidance, professional reviews, and a user-friendly experience without paperwork or consultation fees.

OPPORTUNITY / CHALLENGE

What challenge were you trying to address with Forecasta?

We needed help with AI automation development.

SOLUTION

What was the scope of their involvement?

Forecasta handled two engagements for us. The first was a two-week AI opportunity audit where they mapped our intake-to-filing process, evaluated the value and risk of automating each step, and created a prioritized automation roadmap.They did an excellent job mapping our entire intake-to-filing process and evaluating each step based on two key factors: the value automation could create and the regulatory or reputational risk it could introduce.They delivered a clear, sequenced roadmap that showed us what we should automate, what required additional safeguards, and, importantly, what we should not automate.One of the most valuable findings was that our eligibility criteria were scattered across spreadsheets, saved prompts, and internal knowledge, with no single up-to-date version or historical record. They identified this as a critical issue and explained why we couldn’t safely build automation on top of it until the criteria were properly centralized and structured.That audit led directly to the second engagement, which focused on building a versioned eligibility rule base and an internal assistant. The main deliverables were a centralized versioned ruleset, a test suite, the ability to replay historical eligibility decisions, an internal plain-language assistant, an editing and review process for our team, and the supporting documentation.They also identified another major opportunity that we hadn’t previously built: deadline and expiry tracking, including permit expiries, biometrics windows, and medical exam validity. This was ultimately identified as one of the highest-value, lowest-risk areas for automation.

What is the team composition?

We worked with one teammate from Forecasta.

How did you come to work with Forecasta?

We found them through an online search and a referral.

How much have you invested with them?

We spent less than $10,000.

What is the status of this engagement?

We worked together from July–August 2026.

RESULTS & FEEDBACK

What evidence can you share that demonstrates the impact of the engagement?

The clearest measurable outcome was that our ability to replay past eligibility decisions went from 0%–100%. Before Forecasta's work, if we needed to review a decision from eight months ago, we could only compare it against the current eligibility criteria, even though those criteria may have changed since then.After the project, we could reconstruct any past decision using the rules that were actually in force at that time, with the relevant rule version and supporting evidence. This gave us a much more reliable and auditable eligibility process.Overall, Forecasta gave us much more than an audit. They gave us a clear roadmap for what to build, what not to build, and how to approach automation responsibly. The work was extremely valuable and gave our team a much clearer direction.

How did Forecasta perform from a project management standpoint?

They delivered on time. We communicated via virtual meetings and email.

What did you find most impressive about them?

They had fast communication and a deep understanding of the problem.

Are there any areas they could improve?

No, there weren’t any.

BACKGROUND

Introduce your business and what you do there.

I’m the VP of operations for RentalHelp.ai, a real estate company. We connect tenants and landlords across Ontario. We collect rental listings from three major sources, about 188,000 Ontario listings, and that data is the core of everything we offer.

OPPORTUNITY / CHALLENGE

What challenge were you trying to address with Forecasta?

Our listing data wasn’t reliable. We collected data from three major sources, and the same apartment often appeared two or three times. Some sale listings were being counted as rentals, which put a $2,199 one-bedroom on our homepage. Place names were inconsistent, with Portugal and Florida in our list of Ontario cities. We had no record of when listings came down. We wanted to build AI features on our data, but we couldn’t trust it enough to do that.

SOLUTION

What was the scope of their involvement?

Forecasta built a daily pipeline that collected listings from all three sources, merged duplicates so each apartment appeared once, filtered out sale prices and bad values, standardized town and neighborhood names, and recorded each listing’s history: when it was posted, every price change, and when it was taken down. On top of that clean history, they built a fair rent estimate for any unit, a prediction of how long a listing would take to rent at its price, and a monthly rent report by city and neighborhood.

What is the team composition?

We worked with 2–5 teammates from Forecasta.

How did you come to work with Forecasta?

We found them through a referral and Clutch. We chose them because their pricing fit our budget, they were a great culture fit, they offered good value for the cost, and they were referred to me.

How much have you invested with them?

We spent around $10,000–$50,000.

What is the status of this engagement?

We worked together from May–August 2026.

RESULTS & FEEDBACK

What evidence can you share that demonstrates the impact of the engagement?

The clearest evidence is what’s no longer on our site. The $2,199 one-bedroom is gone, and so are Portugal and Florida. Three separate feeds of about 188,000 Ontario listings are now one deduplicated record that updates itself every day, so each apartment is counted once. For the first time, we have a history for every listing: when it went up, when the price changed, and when it came down. That history is what made our three new products possible: the fair rent estimate, the time-to-rent prediction, and the monthly rent report. Before this, we couldn’t have built any of them.We’ve removed more than 5,000 duplicates and established an automated data pipeline and cleansing. Now, our users have higher confidence in our data. Our conversion and engagement rates have increased. Our engagement rate has improved by 15%, and our conversion rate has improved by 10%, from 1% to 1.1%.

How did Forecasta perform from a project management standpoint?

They delivered Phase 1 before starting Phase 2, so we were never building on data we didn’t trust. We had regular check-ins where they showed us real examples from our own data, such as specific duplicates they had merged and specific listings they had filtered out. That made it easy to confirm the cleanup was right. The project finished on the timeline we agreed on.

What did you find most impressive about them?

They cared about whether the numbers were right, not just whether the model ran. They went through the messy cases one by one: the same unit listed on three sites with slightly different details, a listing taken down and reposted, and a sale price disguised as a rent. They were also honest about what the data couldn’t support yet. They didn’t oversell the time-to-rent model. They built it to show how confident each estimate was.

Are there any areas they could improve?

I would’ve liked an earlier conversation about how much history the time-to-rent model needed. Since our sources never recorded when listings came down, that history only started building when the project began. We learned partway through that the estimates would be broad for smaller towns and less common unit types at first. The model handles it honestly, but knowing that from day one would’ve helped us plan how we present the feature to users.