But by analyzing layouts, we unlocked their secrets, detecting multi-columns and preprocessing images to make them OCR-friendly. Custom parsers became our tools of choice, coupled with layout analysis that split sections into readable pieces of data.
We adapted Puppeteer, disguised in stealth mode, navigated like a human. User agents were shuffled, cookies expertly managed, and our Cheerio selectors learned to flex with every small HTML change. Constant monitoring kept us one step ahead of the curve.
Through the power of lemmatization and contextual windows, we unraveled ambiguities. Synonyms had their day thanks to WordNet, while custom industry-specific ontologies brought precision to skill interpretation like never before.
We reshaped the conversation. With sharper prompts, clear contexts, and detailed examples, responses aligned with our expectations. A cache of frequently used outputs took the pressure off the API, and default templates stood ready to fill the gaps.
Encryption became our first line of defense, securing data both at rest and in transit. Role-based access controls narrowed exposure, and every policy was made transparent to users, earning their explicit consent with clarity and integrity.
Initially we implemented Google serper API which searches for the respective mentor based on the roadmap, but sometimes we didn’t get the expected result as it provides random mentors, to overcome this challenge we have scraped the various website eg, websites like mentorcruse and many relevant from where we can get the mentors data (their website link, name, etc) and we store that in database, basically we have created a dataset of mentors and then using tf-idf we have searched the database to provide the relevant mentor, if the skill of the user is different and we do not have that data in our dataset, in that case, we recommend the mentors from youtube.
The system primarily supports English-language resumes and profiles. Users with documents in other languages face reduced accuracy in parsing and analysis.
Limited ability to assess soft skills due to reliance on textual data. Potential underestimation of a user's interpersonal and leadership abilities.
Dependence on external AI services like OpenAI can introduce latency and unpredictability. May affect user experience due to delays in processing and generating content.
Scraping data from platforms like LinkedIn may conflict with their terms of service. Risk of legal action or service denial if not managed appropriately.
Difficulty in accurately parsing resumes with unconventional formats, heavy graphics, or infographics. Important information might be missed, affecting the completeness of the skill analysis.
Processing complex parsing and AI analysis in real-time can strain resources. Scalability issues during high user load periods.
Reliance on user-provided data which may be incomplete or outdated. Inaccurate skill assessments and recommendations.
Train proprietary models (e.g., fine-tuned BERT) for tasks like NER, skill extraction, and semantic analysis. Reduce dependency on external APIs for core functionalities.
React.js: Building dynamic user interfaces for data visualization.
Node.js with Express.js: Server-side scripting and API development.
MySQL: Storing structured data like user profiles, skills, and job roles.
Tesseract.js: Optical Character Recognition (OCR) for extracting text from PDFs.
OpenAI SDK: Leveraging GPT models for natural language understanding and content generation.
Cosine Similarity Libraries (e.g., natural, ml-distance in Node.js): For semantic comparison of skill sets.
