The revealing document in finance is rarely the finished one. A final pitch deck arrives looking immaculate and faintly inevitable: every comparable chosen, every claim trimmed, every chart behaving itself. Yet the intelligence is often in the untidy version that came before. It lives in the comment that killed a weak argument, the cell a vice president recut, the sentence a partner softened because a client would hear it differently. The polish is the result. The redline is the mind at work.
Samir Dutta has built his career around that distinction. Before he was the co-founder and CEO of Farsight, he studied computer science at MIT, worked in software engineering at Yelp, spent three years in Evercore's mergers and acquisitions group, and evaluated technology companies as a growth equity investor at General Atlantic. He learned how software gets made, how deals get presented, and how investors decide what deserves capital. Farsight is where those lessons meet.
The company's premise is plain enough to fit on a Post-it: people hired for judgment should not spend their best hours forcing logos into alignment. Farsight generates the artifacts of financial work - models, pitch books, confidential information memoranda, research, and investment memos - inside Excel, PowerPoint, and Word. The more interesting ambition is to make those artifacts resemble the work of a particular firm, complete with its standards, prior decisions, and preferred way of making a case.
A founder assembled from three jobs
Founder stories tend to tidy themselves in hindsight. Every detour becomes destiny, every summer job a prophecy. Dutta's route needs less embroidery than most. Software engineering supplied the instinct to build. Evercore supplied the production grind. General Atlantic supplied the buyer's eye. If you wanted to design a person for the problem Farsight addresses, you might produce something close to this sequence, though the design would look suspiciously convenient.
The foundation came earlier. At Suffern High School in New York, Dutta graduated as valedictorian and earned National Merit recognition. At MIT, he completed bachelor's and master's degrees in computer science. His public portfolio from those years includes machine-learning projects for music recommendation and predictive resource allocation. He also co-founded WeHear, an earlier technology startup. The young engineer was already interested in systems that take messy signals and produce a useful decision.
MIT also supplied the relationships behind Farsight. Dutta's co-founders, Noah Faro and Kunal Tangri, are fellow alumni. Their experience was complementary: Faro brought engineering work from Amazon; Tangri brought machine-learning experience from Hugging Face and MIT CSAIL. An early Farsight investor met Dutta through the Phi Beta Epsilon fraternity network. It is an origin with less garage mythology and more institutional density - three technical friends, a shared vocabulary, and a problem one of them had encountered at close range.
The expensive theater of making it look easy
High finance contains a peculiar mismatch. Its firms compete on ideas, relationships, and nerve, but much of their junior labor goes into repetitive production. Numbers move from database to spreadsheet. Spreadsheets feed charts. Charts enter presentations. Presentations are reformatted, checked, revised, and checked again. A page that looks effortless at nine in the morning may contain a small archaeological site of midnight decisions.
Farsight's own funding announcement joked that junior finance professionals should make sure their benefits include premium vision. The joke works because the screen hours are familiar. Dutta's sharper point is economic: the cost is not limited to tired eyes or crooked logos. Every hour spent recreating a standard analysis is an hour not spent interpreting it, speaking with a client, or finding the unusual fact that changes a deal.
The market answered with more than polite interest. Farsight reported that revenue rose tenfold in 2024 while its customer roster grew fivefold. By June 2025, it had announced $16 million in total capital, with a Series A led by SignalFire and participation from RRE Ventures, Link Ventures, K5 Ventures, and strategic angels connected to firms including Blackstone, Oaktree, Searchlight, and Bank of America. Several of the people backing the product came from the world expected to use it. That is not proof of inevitable success; it is evidence that the irritation travels.
The product hiding in the markup
Automation alone is a modest target. A general model can draft prose, summarize a transcript, and fill a slide. Dutta's thesis begins where the first draft ends. Junior bankers, he argues, learn from the markup that comes back from a senior colleague. One comparable stays; another disappears. A confident sentence acquires a caveat. The edits encode taste, risk tolerance, client knowledge, and years of pattern recognition. Then the final file goes out and that path vanishes.
Farsight calls the larger idea a “system of judgment.” Rather than treat a finished document as the whole truth, the software is meant to learn from review decisions and approved redlines. The phrase is grand; the raw material is wonderfully ordinary. It is the diff between drafts, the comments in the margin, the recurring correction that eventually becomes a house rule. Engineers have always kept version histories. Finance, in Dutta's telling, has been throwing away a valuable one every evening.
This also explains why Farsight lives inside familiar applications. A separate chat window can produce an answer; it cannot easily inhabit the chain of formatting, formulas, references, comments, and approvals through which financial work becomes sendable. Dutta has compared chat-first AI with the BlackBerry at the dawn of the touchscreen: important, useful, and possibly attached to the wrong interface. His preferred future is less theatrical. The software appears where the work already is.
From tool to infrastructure
After the financing, Farsight began widening the surface of that idea. In August 2025 it acquired Presentable AI, adding technology for natural-language editing of free-form PowerPoint slides. In May 2026 it launched Freeform, designed to turn a prompt and a firm's materials into complete, multi-page deliverables across presentation, spreadsheet, and document formats. The aim was not a clever sample slide. It was the whole unwieldy thing, with a consistent argument and fewer rounds of repair.
The larger test arrived in July 2026. S&P Global Market Intelligence announced a strategic partnership with Farsight and took a minority investment. A planned Capital IQ Pro add-on would pair S&P's data with Farsight's ability to produce firm-specific materials. Integrations with PitchBook and North Data followed in August, bringing private-capital information and European company records into the workflow. The company that began by attacking production work was moving closer to the data infrastructure underneath it.
Scale changes the nature of the claim. It is one thing to make a deck for a friendly early customer. It is another to operate within the habits, security expectations, and quality bars of investment banks and private-equity firms. Farsight's founders have emphasized reliability over spectacle. In finance, a dazzling error is still an error, and often one with a managing director attached.
Count the minutes after the model stops
Dutta's recent writing has become more exact about measurement. Comparing AI products by token price, he argues, ignores the expensive portion after generation: source checks, tie-outs, analyst review, VP comments, and rework. A model that saves twenty dollars but creates forty-five minutes of senior review is not cheaper in any practical sense. His proposed unit is the fully loaded cost per approved deliverable.
This is the banker in the founder speaking. Benchmarks are abstractions; approval is an event. A pitch deck must support its claims and match house style. A financial model must cross-foot. Diligence must lead back to traceable evidence. The job is not finished when the model stops typing. It is finished when a person with something to lose agrees to send the work.
It is also where Dutta's ambition becomes visible. Farsight is not merely trying to spare analysts a few weary hours. It wants each review to improve the next starting point, allowing judgment to accumulate rather than walk out with the person who supplied it. That aspiration comes with hard questions about accuracy, governance, and whether taste can be captured without flattening it. Those questions are not detours from the product. They are the product.
Dutta has gone from writing code to assembling deals to evaluating companies, and now to building one. Each role taught him to notice a different kind of work. The engineer sees a repeatable system. The banker sees the deadline. The investor sees whether the advantage compounds. In Farsight, he is betting that the smallest edits can answer all three.