Albert Chalom Is Teaching AI to Read the Fine Print

The most revealing sentence Albert Chalom wrote about a $54 million financing round concerned a broken heater. PermitFlow, where he is Founding AI Engineer, had just announced its Series B. The conventional ingredients were all present: investors, growth, jobs, capital. Chalom went looking for the things on the other side of the paperwork. He cared about more homes, more solar panels connected to the grid, more EV chargers, and people getting their heat repaired when it failed in winter. The raise mattered because the work could move.

He also wanted, he added, to liquidate his equity someday. This is how one keeps a mission statement from floating off like a party balloon. In Chalom's telling, ambition and a small joke share the same desk. PermitFlow may speak the contemporary language of AI agents. The objects he names are roofs, plugs, wires and radiators. His unit of meaning is something you can bump into.

That gives his career an appealing shape. He went from theoretical computer science at Harvard to collaborative software at Asana, then into a trade whose grandest projects may wait on a form that was designed before anyone thought to put a browser in a pocket. The elegant system met the municipal counter. Chalom chose the counter.

A theory class with muddy shoes

Chalom attended Harvard from 2015 to 2019, earning A.B. and S.M. degrees. His public college record has a charming split personality. On one side sits CS 121, Introduction to Theoretical Computer Science, where he served as a teaching fellow in 2018. On the other sits the Harvard University Band, an organization for which he served as a manager. Proofs and brass, abstraction and logistics: it is difficult to imagine better rehearsal for software engineering.

Cambridge · 2018

Before municipal rules became his problem, Chalom helped students reason about computation itself.

Theoretical computer science is interested in what problems can be solved and what resources a solution demands. Permitting offers a less tidy exam. Every jurisdiction may alter the requirements. A missing document can restart the clock. An inspection is both an administrative event and a human appointment. Information arrives in PDFs, portals and phone calls, each confident that it is the center of the universe.

The teaching work also matters for a simpler reason. Engineering is partly the art of explaining a complex system without lying about it. Students expose every shortcut in an explanation. So do building departments.

Five years inside the work about work

Before PermitFlow, Chalom spent roughly five years at Asana. He had already interned there in 2017, then interned at Quora in 2018. After college he returned to Asana as a full-time software engineer. The company publicly credited him among the people who contributed to its dashboards feature, a tool designed to turn the scattered facts of a project into a legible picture.

There is a thread here. Asana organizes the commitments made by a team. PermitFlow organizes commitments made among builders, documents, agencies and calendars. Both products are interested in coordination, though construction adds a sharp twist: the work eventually leaves the screen. Concrete cures. Crews wait. The weather refuses to read the roadmap.

The move also traded one kind of legibility for another. A workplace product can often observe its own success: a task moves, a chart updates, a teammate receives the message. Construction software hands its result to a longer chain of people and institutions. The permit is followed by a site, a crew and an inspector. Progress may be visible in the software long before it is visible from the street. An engineer has to care about both views without confusing them.

Chalom joined PermitFlow in October 2024, initially described publicly as a founding engineer and now as Founding AI Engineer. “Founding” in this title does not make him a company founder. PermitFlow was started in 2021 by Francis Thumpasery and Samuel Lam. It does locate Chalom near the beginning of the company's deeper AI chapter, where products described as agents take on pieces of permitting, inspection and licensing work.

The paperwork has a physical shadow

PermitFlow's proposition begins with an irritation familiar to builders and obscure to almost everyone else. Construction rules are intensely local. Requirements hide across municipal sites. Applications repeat information. Status updates travel through portals and calls. Each task looks manageable alone; together they form an industry of waiting.

AI fits this terrain because the work blends retrieval, document interpretation, data entry and follow-up. It also resists the fantasy that one model can simply know everything. A credible system needs the right local rule, the current version, the correct project detail, and a way to show its work. PermitFlow says it has assembled millions of permitting data points to support that machinery. The engineering problem is less like asking an oracle and more like running a very alert clerk's office.

That distinction is important in a field fond of theatrical demonstrations. A permit application cannot merely sound plausible. A parcel number, contractor license or setback measurement is either right or capable of making expensive trouble. The useful AI system therefore needs habits that look almost old-fashioned: check the detail, preserve the trail, notice the exception, ask when confidence runs out. Intelligence is welcome. Reliability gets the appointment booked.

Chalom's public comments keep returning to the downstream result. Solar panels are useless if they cannot connect to the grid. Housing plans do not shelter anyone. An EV charger in a proposal deck charges no car. Permit approval is rarely the reason a project exists, but it can become the reason the project stalls.

A memory named Dignity

Long before construction AI, Chalom had already shown a habit of following systems to their human consequence. In 2015, as a senior at Livingston High School in New Jersey, he spoke at a student program modeled on TED Talks. His subject was a mission in Ghana and the children he had seen working on fishing boats.

One image stayed with him: a small boy shrinking while a master recited his faults. The boat beneath them was named Dignity. Chalom brought that bitter detail home and told it to his peers. The story belongs to a very different chapter of his life, and it should not be pressed into a tidy origin myth. It does reveal an eye for the distance between a word and the reality underneath it.

That same distance appears in construction, in a gentler but persistent form. “Approved” may mean a crew can begin. “Pending” may mean a family keeps waiting. Software tends to compress reality into labels. Chalom's best public explanations reverse the process. He opens the label and points to what is inside.

The annex and the billboard

Startup growth supplies its own comic scale. When a PermitFlow designer shared the Times Square advertisement announcing the Series B, Chalom replied that it was exciting to see how far the company had come from an old office annex with a “never normal temperature.” From thermostat trouble to an illuminated Manhattan rectangle: funding announcements have been built from thinner material.

His comments around the company offer other small clues about how he works. He called product leader Kevin Chen his daily brainstorming partner. He praised PermitFlow's ReconstructLA effort, which offered permitting help for rebuilding after the Los Angeles wildfires, and invited people who might benefit to pass it along. The public persona is compact: a colleague's colleague, more interested in extending the work than narrating himself.

The title Founding AI Engineer can sound as if it belongs on a chrome nameplate from the future. Chalom's version comes with municipal websites, construction schedules and the occasional phone call. This is AI in work boots, even if the engineer is still at a keyboard.

What happens next will be measured in ordinary friction. How many details can software extract correctly? How often can a team avoid resubmission? How quickly can an inspection be arranged? PermitFlow's financing gives the company room to hire and build. It gives Chalom more of the same unglamorous question he chose in 2024: what stands between an intention and a finished thing?

There is a pleasing modesty in choosing the obstacle instead of the monument. Nobody poses for a photograph beside a correctly routed application. Few children announce that they want to grow up and reconcile jurisdictional requirements. Yet a city is partly made from such invisible competence. Every finished room carries a long paper shadow, and someone had to shorten it without losing what the rules were meant to protect.

The answer is sometimes code. Sometimes it is a rule written by a town. Sometimes it is a person waiting for heat. The useful engineer learns to see all three.