Along the way, I’ve noticed a recurring pattern: most engineers joining VC come from the outside, and there’s no real onboarding guide for the role.

They’re talented people who haven’t worked in venture capital before, don’t yet know how funds operate, and often spend their early months figuring out the basics: what data matters, what tools exist, where to focus first. I was one of them back in the day.

So I wrote the guide I wish I’d had: Building for Venture Capital.

Subscribe now

What It Covers

The guide is organized into three parts:

Part 1: Understanding VC

How venture capital funds actually work, from the perspective of someone building technology for them. Fund structures, LP relationships, investment processes, and the common mistakes technical hires make when they don’t understand the domain.

Part 2: The VC Tech Stack

A tour of the tools and systems VC funds use: research platforms, sourcing tools, CRM and deal flow, fund operations, portfolio support, fundraising, and external presence. For each category, I cover what the tool does, when to build vs. buy, and real examples from working funds.

Part 3: Technical Foundations

The deeper technical work: choosing your technology stack, working with data providers, data modeling for VC, entity resolution, data quality, warehousing, integrations, security and compliance, and emerging trends like MCP and AI agent orchestration.

There’s also a resources section with curated newsletters, books, research papers, and a prompt bank for common VC workflows.

Why Open Source?

At Inflection, we believe the VC industry benefits when funds share knowledge about how technology is shaping the way we operate. Too often, this expertise stays siloed within individual firms. By publishing this guide, we want to create more conversation about how new technologies are changing what's possible for VC funds of all sizes.

The whole thing is on GitHub. If something’s wrong or missing, you can open an issue or submit a PR. If you’ve got prompts, resources, or perspectives that would help others, contributions are more than welcome.

Acknowledgments

This wouldn’t exist without the people I’ve worked with at Inflection and EQT who shaped how I think about this work. Special thanks to the external contributors who shared their perspectives.

Check it out at buildingfor.vc. Feedback welcome.

// AP

Subscribe now

The Kepler space telescope was NASA’s first planet-hunting mission, assigned to search a portion of the Milky Way galaxy for Earth-sized planets orbiting stars outside our solar system.

- NASA

Doing Our Homework

At Inflection, we really enjoy “getting into the weeds” of emerging ideas and industries. It’s part of our firm’s DNA, what leads to outsized returns, and what makes the job of investing interesting!

You can dive into our published research here.

An Overlooked Need

Part of our thesis around Data-Driven VC is that building the future of venture capital requires tooling across the value chain, not only in sourcing.

We’ve seen how tools such as Perplexity are fantastic for guiding research, and tools such as Notion are great for writing and collaboration, but through our experience noticed these drawbacks for investment research:

• Limited Collaboration: AI research assistants like Perplexity and ChatGPT are fundamentally single-user tools. You can share links to chats you've had, but there's no way to "jump-in" and collaborate on exploring a topic together.

• Fragmented Workflow: The current research process requires constant switching between multiple tools – AI assistants for initial exploration, research papers on arxiv.org for deep dives, and separate tools for synthesizing findings. This context-switching makes it harder to maintain a coherent investment thesis and wastes valuable time.

• Generic Intelligence: Existing platforms are designed for general users doing broad research. Venture investing requires specialized capabilities – like mapping emerging technology landscapes, tracking shifts in market dynamics, and building deep expertise in specific domains. Getting meaningful insights from these existing tools requires careful prompt engineering or manual synthesis, which slows down the team.

We weren’t able to find an out-of-the-box tool that addresses these pain-points, so we decided to build something ourselves: Kepler.

The Kepler Platform

Mission Goals

• Automate Research Generation: Streamline the process of collecting, synthesizing, and distributing research on emerging trends, enhancing the fund's ability to identify new opportunities quickly.

• Enhance Proactive Research: Conduct more comprehensive research earlier in the investment process to reduce time between identification of a potential investment and making an investment decision.

• Expand Pathfinder's World View: Add research on emerging industries identified in Kepler to Pathfinder (our sourcing engine) to increase the volume of high-quality information available to Pathfinder, enabling more informed assessments of companies and founders.

Key Features

Explorations

At the heart of Kepler is a chatbot experience powered by agents specialized for VC research. Unlike traditional research tools, Kepler enables true collaboration, allowing investment team members to work simultaneously within the chat interface. The platform leverages existing context from research notes, uploaded documents, scraped websites, agent memories and tools to provide richer insights. Sources are cited so the team can verify the accuracy of information in the agent’s responses.

Users can "fork" explorations to pursue new directions while maintaining the original thread, and with a single click, synthesize their explorations into structured research notes. This seamless flow from exploration to documentation keeps insights from getting lost in the research process.

Research Notes

Research Notes in Kepler combines a familiar, rich editing experience (similar to Notion) with the power of an AI assistant that has access to all of Kepler's accumulated knowledge.

This creates a fluid experience where the assistant can reference any piece of research, data, or insight the team has previously captured while helping craft new analyses.

Additional Sources

Kepler's knowledge base grows in two ways: users can directly upload documents or scrape websites themselves, and the platform automatically captures new sources during exploration. When Kepler's agents cite a source, that source is automatically added to the knowledge base – whether it's a research paper, website, or industry report.

This means the platform becomes more intelligent and context-aware with every interaction, building a rich corpus of institutional knowledge that grows both intentionally and organically through the team's research activities.

Impact

Like its namesake – NASA's space telescope that systematically hunted for habitable planets – our Kepler systematically hunts for emerging industries and trends that could power the next generation of transformative companies.

While Kepler is just two weeks into its launch, we expect three key areas of impact:

• Faster iterations on research through purpose-built agents that understand our firm and investing strategy

• A more aligned investment team through collaborative research experiences

• Growing corpus of institutional knowledge that compounds with every interaction

Where We’re Heading

Kepler represents just the beginning of our journey to revolutionize research tools at Inflection. Our next phase of development will focus on:

• Building a comprehensive knowledge graph to connect insights across our firm

• Sharing our research through public-facing "Inflection Points"

Interested in learning more about Kepler or our other analytics and automation tools? Please reach out to ap@inflection.xyz.

Onwards & upwards

Team Inflection

Last week, Inflection wrote a guest article on Dr. Andre Retterath’s Data-Driven VC newsletter:

Data-driven VC

6 Lessons From Leveraging Data & AI as a Micro VC Fund With $50M AUM

👋 Hi, I’m Andre and welcome to my newsletter Data-driven VC which is all about becoming a better investor with data & AI. Every Tuesday, I publish “Insights” to digest the most relevant startup research & reports. Every Thursday, I publish “Essays” that cover hands-on insights about data-driven innovation & AI in VC, and every…

Read more

2 years ago · 15 likes · Andre Retterath and Alex Patow

Andre’s done a tremendous job of building a community of investors, engineers, data scientists and generally curious people who are interested in learning more about leveraging data in VC (over 25,000 strong 🤯). You can discover more content on his site.

As he eludes to in his intro, we didn’t see eye-to-eye on one of his recent posts, “Can We Fully Automate Startup Investing?”. This led to a great conversation and, ultimately, this article.

Since publishing, there’s been an outpouring of people interested in starting something at their firm or swapping “war stories” from their own data-driven initiatives. It’s incredibly energizing.

Below is the article in full. If you’re curious for more, please reach out.

Data-Driven Investing as a Micro Fund

At Inflection, we firmly believe that the future of VC is “data-driven”, especially for funds our size. We also equally value the human element of venture capital. There's no substitute for deep connections with founders, other VCs, and our own intuition. We believe this combination of data and human insight is where outsized returns become possible.

Being data-driven, for us, means building tools that augment the entire team's work, not just that of investors. We focus on enhancing all aspects of our operations, not only sourcing investments. As the sole "data person" in our six-member team, my role involves wearing many hats, but ultimately, my mission is to scale the efficiency and impact of our team through strategic use of data and software.

Second Wave of Data-Driven VC

We’re now entering our second wave of “data-driven VC”. Early pioneers in this space (such as EQT, Earlybird, Moonfire, amongst others) had to build substantial infrastructure from scratch:

• Complex pipelines to wrangle spotty and immature data sources

• Training and hosting custom models

• Creating platforms due to a lack of VC-specific tools (particularly CRM systems)

While there's still value in prioritizing in-house development, it requires a large team to implement effectively (something that’s not realistic for funds of our size).

Fortunately, the industry has evolved significantly since the first wave:

• Proliferation of (relatively) stable, cheap, and reliable Large Language Models (LLMs)

• Decreasing cost and increasing availability of out-of-the-box data and tools, specifically targeted at our industry (we’re big fans of Specter, Gravity, People Data Labs, and Attio for our needs)

• Increasing engineering efficiency through AI-assisted coding, data transformation, and cloud deployment tools

We expect these trends to accelerate, with advancing technologies enabling even smaller funds to compete effectively, shifting the key differentiator from data infrastructure to insight interpretation and action.

To funds looking to become data-driven: the time is now.

Our First Year

Inflection’s initial blog post on “An Engineering Approach to Venture Capital” was published just over a year ago. I started with the firm in November of last year.

Since then, we’ve built some really cool stuff, we’ve also done some really “boring” work:

• Launched our internal deal sourcing tool, Pathfinder

  • Utilizes internal and external data, LLM agents, and a graph database for sourcing opportunities in our sector

  • Automatically populates our CRM with sourced opportunities for the investment team's review

  • It’s currently evaluating roughly 500 new companies and 4000 founders per week, a task that would be impossible for us without tooling

• Brought in best-in-class tools for deal tracking and portfolio hiring

• Reduced the time for creating valuation memos for our audit process by 80%

  • Achieved through LLM Agents and Retrieval-Augmented Generation (RAG)

• Acquired signal and people data sources to support our projects

• Refreshed our website

• Streamlined our toolset by eliminating unused tools

Like I said, not everything is glamorous, but it’s important for being a holistic, data-driven fund.

Lessons Learned

Although we are early on in our journey, we learned some lessons along the way:

#1 Start Small, as Small Projects Snowball

While building our data-driven sourcing tool (Pathfinder), we were asked by the operations team to help automate the creation of this year's valuation memos. These letters summarize the investments we’ve made, their current valuation, and the rationale for their valuations for the audit process. As there’s one memo for each investment, it’s typically a time consuming task to complete manually. By doing a small, one-week project we were able to validate Pathfinder's tool stack (specifically LLM Agents) in a controlled environment.

These learnings significantly accelerated our progress on sourcing. Importantly, it also helped achieve buy-in from our operations team on the vision of a data-driven fund. We've found that starting with small, manageable projects can lead to unexpected benefits and growth for longer-term strategic projects.

#2 Take Calculated Risks

The first iteration of Pathfinder relied on a linear process where LLM agents gathered the information needed to evaluate potential investment opportunities as they arrived in the system. While functional, this approach had limitations in providing context and integrating existing results, often leading to confusing outcomes. Recognizing these limitations, we decided to take a calculated risk by rethinking our data architecture and transitioning to a graph database.

We believed that modeling our database to mirror the entities and relationships investors use in assessing companies would lead to better results from our LLM agents. This required substantial rework, but the results have been significantly more aligned with our investment thesis and early-stage opportunities. Adopting a GraphRAG approach has enhanced the agents' contextual understanding (see Neo4j's writeup on GraphRAG, and stay tuned for more on our blog). This decision exemplifies how embracing calculated risks, when guided by clear objectives, can drive innovation and unlock new value.

#3 Embrace Nimbleness

Some tools we build may only provide 3-6 months of “alpha” before external tools emerge and commoditize that part of our platform. When more mature, off-the-shelf solutions become available, we transition from in-house development to leveraging these external solutions. This approach allows us to redirect our focus and resources toward identifying and developing the next source of alpha.

#4 Secure Buy-in from the Team

Team support is crucial when implementing data-driven initiatives at a VC fund. Achieving reliable, scalable, and impactful results takes time, and not all efforts will succeed. Organizational buy-in enables experimentation and grants the independence needed for effective decision-making.

At Inflection, the team had already explored various Python scripts and external tools to enhance workflows, from podcast transcription to research summarization and web scraping. Recognizing the potential leverage of a technical role in the post-LLM era, the investment team outlined initial ideas on their blog.

To retain this alignment and maintain momentum, we implemented a structured approach:

1. A week-long kick-off session to set principles and direction

2. Weekly product calls and bi-monthly in-person work days

3. Bi-annual presentations on innovations and roadmaps

This structure fosters tight feedback loops and a culture of innovation.

#5 Find a Product-Engineering Hybrid

Look for someone who enjoys both product and engineering. In smaller funds, this role often falls to one person. Much of the job involves managing product, not just engineering in isolation. It's essential to have someone who can excel (and enjoys working!) in both areas to ensure what's being built matches the fund's needs. This should be a full-time role for someone passionate about building, not a part-time job for an investor. 

#6 Develop Guiding Principles, Not a Backlog

Start by establishing guiding principles for the first 12 months, rather than specific projects. Let the projects emerge over time. For us, this meant creating an architectural philosophy that addressed questions like how to approach "buy vs. build", how to balance "speed of iteration vs. quality of output", and whether to focus on "platforms" or "projects". This approach gives us the comfort to get started without locking us into specific projects or implementation details, allowing for flexibility as we learn and grow.

What’s Next for Inflection

• Continuously improving our sourcing tool, Pathfinder

  • Adding new data sources (research papers, social media, etc.) and giving our agents ability to expand the graph through their “intuition”

  • Crafting a “magical” experience for the investment team to input and extract information from our knowledge graph (we love the work USV has done with their “Librarian” tool)

• Generate novel ideas for companies and areas of exploration based off of research and our thesis

• Launching tools for our founders that go beyond just hiring talent

• Actively growing the community of data-driven VCs by:

  • Open sourcing our work

  • Collaborating with our community and new data providers

  • Presenting at industry events

Hopefully this serves as a helpful guide for smaller funds looking to be “data-driven”. We believe that by embracing these practices, micro funds can punch above their weight and compete effectively in today's fast-paced and competitive venture landscape.

Vision: A computer aided venture investing machine (CAVI)

At its core, a VC firm is a decision-making and insight machine, an amalgamation of mental models, frameworks and human intuition. We believe this machine can be enhanced and optimised. We envision a world where VC operations - currently largely manual - transform into an elegant synergy of human and machine, powered by data-centric processes and automation. We called this concept CAVI - computer aided venture investing. This fusion will enable us to leverage our most valuable asset – human attention – more efficiently to focus on non-fungible activities such as relationship building, ecosystem building or creative thinking in dynamic environments with incomplete information.

Here’s what we’re going after: Imagine the first time you held an iPhone or the first time you were recommended exactly that song you wanted to listen to but didn’t know it yet. That is the kind of step function improvement CAVI can bring about in venture capital.

Creating a well-designed decision making system helps to mitigate the influence of personal biases and emotional factors, allowing for more objective and consistent outcomes.

Paraphrasing Ray Dalio

Catalysts

Up until now, it was really hard to be very data-driven and tech-first in venture (as public interviews with some of the larger funds will tell you). The data is inconclusive, mostly it’s people and ideas, without little quantitative support to indicate correlation with returns. Research and contributions have been made by other funds - large and small - going from standardised personality tests of founders to predicting performance by proxy in website visits. We want to build on these efforts and take it to the next level. The exponential advancements in AI, machine learning, NLP coupled with the increasingly ubiquitous access to data have opened up unparalleled opportunities to reimagine the VC landscape.

Those are some of the core drivers paving the way for such shift:

1. Text processing and manipulation is leaps and bounds ahead of where it was just six months ago: The introduction of GPT-4, the latest language model from OpenAI has made it dramatically easier to analyse and extract insights from text data.

2. More data is available than ever before: The digital era has created an explosion of data. This ever-growing wealth of information can be harnessed to provide a much deeper and broader understanding of research models, startups and markets.

3. Emerging venture funds are redefining the playing field: The traditional VC model has its strengths, but it also has limitations. As the startup ecosystem continues to grow and change, our industry must also adapt to remain effective.

In conclusion, the time is ripe for a more data- and automation driven approach to venture capital. By combining the power of advanced technology with the irreplaceable human element, we can create a venture engine that is scalable, efficient and effective in today's fast-paced, data-rich world. If you’re more interested in the broader topic in general we highly recommend following and the community.

A starting point - decomposing the VC value chain

Before we dive into how the venture capital firm of the future could look like more concretely, let’s explore its underlying principles and analyse how the core of its value chain might be affected by the ongoing compute revolution.

This is one out of many ways to define a venture capital firm’s key functions. In [] we defined where the function sits on the spectrum of automation potential where 0 = non fungible human competency and 10 = fully automated. We used a rule of thumb framework taking into account (1) reliability & availability of high quality data allowing for extrapolations into the future (2) required creativity and intuition in highly dynamic environments with incomplete information, (3) required human coordination with external human stakeholders to fulfil the function.

Thematic Research [7]

Where to look at; which hay stack to explore.

Inputs for this function could be

• viability: research breakthroughs, technology maturity, technological limitations like scale, dependencies

• catalysts: convergence, amplification through other emerging technologies, regulatory risk and incentives, demand drivers, geo political and macro economic forces

• impact: degree of novelty, superiority vs. incumbent solutions, societal and economic implications

• market: large problem set yielding multi billion dollar market opportunities within 5-10 years from investing

• popularity: controversial sentiment (we have to be contrarian & right), low investment volumes, low web traffic and media coverage

A shout out to Lawrence and our friends from Lunar Ventures who inspired us through their State of the Future Tracker Methodology.

While some of those inputs can be pulled from research institutions, open source research platforms and the like, qualitative insights might need to be non fungible like expert interviews, hence the 7/10.

Sourcing & Pre-Screening [8]

What to spend time on; which needles in the haystack might be worth analysing.

This function is tricky. If the funnel is too narrow the firm might lose out on potential outliers. If it is too broad it drowns in noise and struggles to focus.

Inputs to this function include:

• network analysis: connections between people; connections between people and organisations; connections between people and ideas / problem sets

Outbound sourcing will be informed by thematic research as we invest with low velocity and high conviction. Our outbound sourcing filters will be set accordingly what can be strongly technology assisted. On the other hand, a variety of other factors will play a rule, such as (1) thematic content creation and events, (2) establishment of a scouting and mentorship network for soon-to-be founders etc. which have more of a human touch. Hence the 8/10 score.

Recommended further reading: More needles, bigger haystacks: What we mean when we talk about data-driven VC.

Investment Decisions [4]

What to invest into; which needles to pick from the haystack.

This function lies at the very core of our business. Over almost a decade we developed frameworks, score cards and various heuristics to take maximally informed decisions optimised for upside potential in the early life cycle of a company.

Inputs to this functions include (over-simplifying here; each of those bullets can be universe in itself):

• Team: personal fit; complementary to existing portfolio founders; shared vision & purpose; deep co-working history; alien skills; obsession with problem space; history of outstanding achievements; leadership that attracts elite talent; resilience and grit

• Product: fundamentally revolutionary approach to solving the problem (0→1); enabling something previously impossible; strong first- or early mover advantage; unique technical moat (network effects; brand)

• PMF: 10x improvement for customers (faster, better, cheaper); high retention; high product velocity; validated demand in target market

• Market: Large ($10BN+) current market and / or growing rapidly; blue ocean or unique approach to gain market share; ability to capture value demonstrated; business logic allows for strong network (ownership) effects and defensible valuation; clear monetisation strategy

• Terms: pre-seed or seed valuation; ownership target of X; 100x upside potential; complementary potential syndicate; 24+ months of runway post financing; clean set up (no previous debt, side deals or skewed incentives

Note: The more early stage (read pre product or MVP stage) a fund operates the more weight needs to be put on research (see above), market and team. Diligence on most of these items can only be automated to a limited degree, hence a 4/10 score for us as a first-check fund. This might look vastly different for later stage funds harnessing much more data (number go up!).

Deal Execution [4]

Pitching our firm to get an allocation; Term negotiations; Due Diligence; Syndicate construction.

This function is one of the more non-fungible ones we can image as it is profoundly related to human relationship and trust building. For the more formal parts including the review of data rooms and legal documents or finding ideal syndicate partners technology might be helpful. Hence, 4/10.

Founder Support & Platform [8]

Strategic Sparring; Board Roles; Founder Coaching; Go to market; Pricing Strategies; Pivots; Key hires; Business Development; Shared Resources; Best Practices.

This function is critical to support our founder ecosystem. Historically, it has been the most non-fungible activity a venture firm’s partners have been involved in (besides fundraising). We are of the belief that this is about to change. Most of the problem spaces mentioned above can be approached through frameworks. The underlying problem spaces are unique in some ways as every company, team and market conditions are different - but at the same time structural patterns remain similar.

• network analytics to identify key hires

• Custom trained LLMs to relay specialised knowledge

• Shared software- and data infrastructure

E.g. every company finding product-market-fit will define modular experiments to test hypothesis and measure success. How this can be done best is specialised knowledge which can be sourced from the public domain and extended through expert-input. Aggregated insights could be packaged in a custom data set and made accessible through LLMs for example.

Artificial agents helping innovators to navigate mission critical challenges will seem like a natural expansion of what venture investors do today. Through data-network-effects (knowledge is a network) such services will improve over time and can provide a competitive edge for an entire ecosystem.

Therefore, 8/10.

Fund Operations [9]

Portfolio Modelling; Investor Reporting; Investor Relations; HR; Legal; Finance; Accounting

These functions are very broad and currently necessitate a lot of human engagement and coordination with external stakeholders.

Inputs to this function include:

• Strategy: portfolio modelling and construction; sensitivity analysis

• Tracking: Portfolio performance tracking; automated quarterly updates; Queries for LP requests

• Audits: Preparation of audit reports; facilitation of audit memos

• Legal: screening & summarising long winded legal documents

• Finance: Cash flow planning; expense analysis

• Fundraising: LP discovery

Given the vast breadth of operations and tooling involved the software and data stack for such functions is typically is fragmented. Hence, the optimisation potential is very high but complex - 9/10.

Putting it all together we can draw an analogy to Hans Moravec’s framework comparing computer performance with water slowly flooding the landscape. We wonder what this landscape will look like in a decade from now… in the meantime we are building Arks!

Computers are universal machines, their potential extends uniformly over a boundless expanse of tasks. Human potentials, on the other hand, are strong in areas long important for survival, but weak in things far removed. Imagine a “landscape of human competence,” having lowlands with labels like “arithmetic” and “rote memorization,” foothills like “theorem proving” and “chess playing,” and high mountain peaks labeled “locomotion,” “hand-eye coordination” and “social interaction.” Advancing computer performance is like water slowly flooding the landscape. A half century ago it began to drown the lowlands, driving out human calculators and record clerks, but leaving most of us dry. Now the flood has reached the foothills, and our outposts there are contemplating retreat. We feel safe on our peaks, but, at the present rate, those too will be submerged within another half century. I propose that we build Arks as that day nears, and adopt a seafaring life!

Hans Moravec, “When Will Computer Hardware Match the Human Brain?” Journal of Evolution and Technology (1998), vol. 1.

Onwards & upwards

Team Inflection