Brahma / build log / thebrahma.dev

Building AI-native products in public · 5 projects tracked

Suggestions

0

To-Do

1

Merge

Purpose

Finding meaningful relationships is difficult, especially for developers whose interests, passions, and personalities are often reflected more accurately in the projects they build than in traditional dating profiles.

Merge is a dating platform built specifically for developers, where GitHub becomes the foundation of compatibility. Instead of relying solely on photos and bios, Merge analyzes repositories, open-source contributions, coding habits, technology preferences, and project interests to create authentic compatibility between people.

The goal is to help developers form meaningful relationships based on shared passions, complementary skills, and genuine compatibility rather than superficial swipes.


Approach

Merge is powered by a collection of AI agents orchestrated through long-running workflows.

When a user joins the platform, agents collaborate to:

  • Connect and securely analyze the user's GitHub profile.
  • Extract technical interests, favorite technologies, and project domains.
  • Understand coding patterns, collaboration style, and contribution history.
  • Generate an AI-powered Developer DNA profile.
  • Create semantic embeddings representing the user's technical interests and personality signals.
  • Match users using a combination of shared interests, complementary strengths, and relationship preferences.
  • Generate compatibility reports and personalized conversation starters.

Temporal workflows orchestrate the entire pipeline, allowing GitHub analysis, AI inference, compatibility scoring, and match generation to execute reliably while providing transparency into every stage of the matching process.


Key Features

  • GitHub-powered profile creation
  • AI-generated Developer DNA
  • Repository and README semantic analysis
  • Technical interest extraction
  • Coding habit and activity insights
  • AI compatibility scoring
  • Shared technology and project matching
  • Complementary personality matching
  • AI-generated conversation starters
  • Detailed compatibility reports
  • Privacy-first GitHub data controls
  • Human-editable AI profile and interests

Tech Stack

Not Decided Yet


Technical Highlights

  • Long-running AI workflows orchestrated through Temporal.
  • Automated GitHub repository analysis and feature extraction.
  • Semantic embeddings generated from repositories, READMEs, topics, and project metadata.
  • AI-generated Developer DNA derived from coding behavior, collaboration patterns, and technical interests.
  • Hybrid recommendation engine combining explicit dating preferences with embedding similarity and complementary compatibility scoring.
  • Explainable compatibility reports highlighting shared interests, complementary strengths, and personalized conversation starters.
  • End-to-end workflow from GitHub authentication to AI-powered matchmaking.

Future Roadmap

  • AI-generated first messages based on shared projects.
  • Voice introductions powered by AI.
  • Co-founder mode for startup matching.
  • Hackathon teammate matching.
  • Friend mode for networking.
  • Local developer community discovery.
  • Event-based matching for conferences and hackathons.
  • Personalized weekly compatibility recommendations.
  • GitHub activity-based profile freshness.
  • AI relationship insights and compatibility evolution over time.
Added Jul 3

In Progress

1

ProspectFlow (Name not decided)

Purpose

Outbound sales teams spend significant time identifying prospects, researching companies, understanding pain points, and personalizing outreach before a conversation even happens.

ProspectFlow automates this entire workflow. Users define their ideal customer profile using filters such as industry, company size, geography, and other targeting criteria. The platform then discovers potential prospects, researches them, builds personalized outreach strategies, and initiates outbound calls through AI voice agents.

The goal is to help businesses scale highly personalized outbound sales without scaling headcount.

Approach

ProspectFlow is built around a collection of specialized agents orchestrated through long-running workflows.

Once a campaign is created, agents collaborate to:

  1. Discover companies matching the target audience.

  2. Research each prospect using public information.

  3. Analyze company context, challenges, and opportunities.

  4. Generate personalized sales angles and messaging.

  5. Create AI voice agents tailored to the campaign.

  6. Execute outbound calls and qualify leads.

  7. Capture conversation outcomes and insights.

Temporal workflows coordinate the entire process, enabling reliable execution across thousands of prospects while maintaining visibility into every stage of the pipeline.

Key Features

  • ICP-based lead discovery

  • Geographic and demographic targeting

  • Automated prospect research

  • Multi-agent workflow orchestration

  • Personalized sales messaging generation

  • AI-powered outbound calling

  • Lead qualification and scoring

  • Conversation summaries and insights

  • Campaign monitoring and analytics

  • Human review and intervention points

Tech Stack

Frontend

  • Next.js

  • TypeScript

Backend

  • FastAPI

  • Pydantic

  • PostgreSQL

Workflow Orchestration

  • Temporal

Voice Infrastructure

  • Vapi

  • Twilio

Infrastructure

  • Railway

Technical Highlights

  • Long-running agent workflows managed through Temporal

  • Multi-stage lead research and qualification pipeline

  • Personalized outreach generated per prospect rather than template-based campaigns

  • AI voice agents dynamically configured from campaign context

  • Reliable execution with retries, checkpoints, and workflow observability

  • End-to-end outbound pipeline from lead discovery to qualified conversation

Jun 2026 → now

Done

3

AskDoc

Purpose

Most document Q&A tools search through chunks of text and generate answers from retrieved passages. While effective, this approach often loses document structure and makes citation verification difficult.

AskDoc explores a different approach: understanding a document through its structure first.

The goal was to build a document assistant that can answer questions while maintaining a direct connection to the original pages, making every response easy to verify.

Approach

Instead of embedding every chunk of a document into a vector database, AskDoc first generates a structured Table of Contents representing the document's topics and page ranges.

When a user asks a question, the AI uses this high-level document map to identify the most relevant sections and pages before loading detailed context. This significantly reduces the amount of document content that needs to be processed while preserving the document's original structure.

Because retrieval happens at the page level, AskDoc can directly link answers back to the source pages and highlight the supporting text in the document viewer.

Key Features

  • AI-generated Table of Contents for every document

  • Page-level retrieval instead of chunk-level search

  • Citation-backed answers

  • Direct navigation to source pages

  • PDF and DOCX support

  • Source passage highlighting

  • Multi-model support through Anthropic, OpenAI, and Google models

  • Context-efficient querying for large documents

Tech Stack

Frontend

  • Next.js 15

  • React 19

  • TypeScript

  • Tailwind CSS

  • react-pdf

Backend

  • FastAPI

  • Pydantic AI

  • Anthropic

  • OpenAI

  • pdfplumber

  • MarkItDown

  • PyMuPDF

  • LibreOffice

Infrastructure

  • Vercel

  • Railway

Apr 2026

ClearCut

Purpose

Most background removal tools require users to upload images to third-party servers, pay per image, or create an account before they can use the product. For quick edits, these requirements add unnecessary friction and raise privacy concerns.

ClearCut was built to prove that modern AI image processing can run entirely inside the browser. Users can remove backgrounds from images instantly without uploads, accounts, subscriptions, or server-side processing.

Approach

ClearCut runs a complete background removal pipeline on the client using the BRIA RMBG-1.4 segmentation model. The model is downloaded once, cached locally, and reused across sessions.

When an image is uploaded, inference runs inside a dedicated Web Worker to keep the UI responsive. The application automatically detects WebGPU availability and falls back to WASM when necessary, ensuring compatibility across browsers and devices.

The generated segmentation mask is applied directly to the image using the Canvas API, producing a transparent PNG without sending any data outside the user's device.

Key Features

  • Fully client-side background removal

  • No image uploads or server processing

  • Privacy-first architecture

  • WebGPU-accelerated inference

  • Automatic WASM fallback

  • Local model caching with IndexedDB

  • Responsive processing through Web Workers

  • Transparent PNG export

  • No accounts, subscriptions, or watermarks

Tech Stack

Frontend

  • Next.js 16

  • React 19

  • TypeScript

  • Tailwind CSS

ML Infrastructure

  • Transformers.js

  • ONNX Runtime Web

  • BRIA RMBG-1.4

  • WebGPU

  • WebAssembly (WASM)

Browser APIs

  • Web Workers

  • IndexedDB

  • Canvas API

Technical Highlights

  • Entire AI inference pipeline runs in-browser

  • 44MB model cached locally after first visit

  • Zero-copy processing using Transferable Objects

  • Automatic backend selection between WebGPU and WASM

  • Static deployment with no backend infrastructure

Apr 2026

Clect.One

Credits: Clect was co-built with Shibam Saha (https://github.com/s4shibam/shib-ai). We collaborated on the product from the ground up, combining AI-powered candidate evaluation with a modern recruiter workflow to create a more transparent and efficient hiring process.


Purpose

Hiring teams spend countless hours reviewing resumes, comparing candidates, and manually evaluating applications against job requirements. Traditional ATS platforms rely heavily on keyword matching, making it difficult to identify strong candidates with unconventional backgrounds or transferable experience.

Clect was built to automate candidate evaluation using AI. Instead of filtering resumes through rigid rules, it analyzes a candidate's profile holistically, scores them against role requirements, and provides recruiters with a structured breakdown of strengths, weaknesses, and overall fit.

The goal is to reduce screening time while helping teams make more consistent hiring decisions.

Approach

Recruiters create a role by defining job requirements, evaluation criteria, and weighting preferences. Candidates can apply through a generated application form and upload resumes in any format.

Once an application is received, Clect extracts and normalizes candidate data, evaluates it against the job requirements, and generates a detailed scorecard. Rather than producing a single score, the platform explains the reasoning behind every evaluation category, giving recruiters visibility into how recommendations are made.

The evaluation framework is configurable, allowing organizations to prioritize different attributes such as technical expertise, project quality, communication skills, leadership experience, or domain knowledge.

Key Features

  • AI-powered resume evaluation

  • Custom hiring scorecards

  • Configurable evaluation criteria and weightages

  • Automated candidate ranking

  • Detailed candidate fit analysis

  • Resume parsing and normalization

  • Application form builder

  • Recruiter dashboard for candidate management

  • Explainable scoring with evaluation breakdowns

  • Candidate comparison across applications

Tech Stack

Frontend

  • Next.js

  • React

  • TypeScript

  • Tailwind CSS

Backend

  • Node.js

  • Express

  • MongoDB

AI

  • OpenAI

Technical Highlights

  • Converts unstructured resumes into structured candidate profiles

  • Evaluates candidates against custom role-specific criteria

  • Generates explainable hiring recommendations instead of keyword-based filtering

  • Supports configurable scoring frameworks for different hiring workflows

  • Designed to reduce manual screening effort while maintaining evaluation consistency

Nov 2025 → Dec 2025