Tech-Santa

Project Name

Tech Santa (Turning Noise into Innovation)

Tagline

An autonomous engine that turns the "noise" of internet complaints (Reddit, Support Tickets) into validated, architected engineering blueprints.

Inspiration

"90% of startups fail. Not because they can't code, but because they build things nobody wants."

We noticed a massive disconnect:

The World Screams Problems: Every day, millions of people post on Reddit, Twitter, and forums saying, "I hate this process" or "Why hasn't someone fixed this?"
Builders Guess Solutions: Founders and engineers sit in rooms brainstorming ideas, hoping they find product-market fit.

We realized the answers aren't in a brainstorm session—they are hidden in the noise of the internet. We built the Tech Santa to act as a "Translation Layer" between human frustration and engineering innovation.

What it does

The Opportunity Compass is a B2B SaaS engine that listens to chaotic textual data and turns it into structured opportunities.

Ingests & Filters: It takes raw user complaints (e.g., "My back hurts from this chair").
Judges Feasibility: It uses an AI Agent to determine if a problem is solvable with current tech (filtering out "I wish I could fly" vs keeping "I wish my chair had better lumbar support").
Deduplicates at Scale: It identifies when 10,000 people are asking for the same thing using different words (e.g., "Cab" vs "Taxi" vs "Ride").
Architects the Solution: It doesn't just find the problem; it generates a Technical Spec, identifying the industry (e.g., HealthTech), quantifying the impact, and suggesting a tech stack (e.g., "Use Python + Computer Vision").

How we built it

We prioritized Technical Depth over simple API wrappers.

Hybrid Search Engine (The Brain): We couldn't rely on simple keywords. We built a custom retrieval pipeline using Gemini Embeddings (Vector Search) weighted at 60% and BM25 (Keyword Search) weighted at 40%. This solves the "Vocabulary Mismatch Problem" (understanding intent vs just matching words).
Idempotency Agent (The Gatekeeper): To prove market demand, we need to count duplicates accurately. We implemented a two-stage process:
1. Fast Path: Cosine Similarity check on vectors.
2. Smart Path: A dedicated Gemini 2.0 Flash agent that acts as a judge to verify if two requests are functionally identical.
Backend: Python FastAPI for high-performance async processing.
Database: MongoDB Atlas to store vector embeddings.
Frontend: Streamlit for the investor dashboard.
Audio: ElevenLabs API to generate "Elevator Pitch" summaries of the opportunities.

Challenges we ran into

The "Context" Trap: Initially, the AI would think "I hate traffic" and "I hate slow internet" were the same because the sentiment was identical. We had to tune the embedding dimensions to focus on topics/nouns rather than sentiment/verbs.
Latency: Running an LLM check on every incoming data point was too slow. We optimized this by using a "Vector First" filter—only calling the expensive LLM when mathematical similarity > 80%.

Accomplishments that we're proud of

The "Curing Cancer" Hypothesis: In our testing, we fed the system 5,000 scattered comments about post-chemotherapy fatigue. The system successfully clustered them and proposed a "Biometric-Syncing Scheduler API" for oncology patients. This proved our tool can go beyond simple apps and find deep, interdisciplinary solutions.
Ranking Algorithm: Successfully implementing the Hybrid Search (Math + AI) formula.

What's next for Tech Santa

Integration: Building a Chrome Extension for Reddit that adds a "Solve This" button, allowing users to submit problems directly to our engine.
Marketplace: Creating a platform where startups can buy validated problem sets ("Golden Tickets") that come with 5,000 verified upvotes of demand.
Real-Time Firehose: Connecting to the Twitter Firehose API for real-time trend detection.